1 00:00:04,821 --> 00:00:05,188 WELCOME, EVERYBODY. 2 00:00:05,188 --> 00:00:09,559 THIS IS MY FINAL MEETING AS 3 00:00:09,559 --> 00:00:15,398 CO-CHAIR -- AS CHAIR OF THE NLM 4 00:00:15,398 --> 00:00:18,135 BOARD OF SCIENTIFIC COUNSELORS. 5 00:00:18,135 --> 00:00:19,436 KING JORDAN IS JOINING ME, HE 6 00:00:19,436 --> 00:00:22,973 WILL BE CHAIRING THE NEXT 7 00:00:22,973 --> 00:00:25,842 MEETING, THE MEETING OF THE 8 00:00:25,842 --> 00:00:27,577 NATIONAL LIBRARY OF MEDICINE 9 00:00:27,577 --> 00:00:32,115 BOARD OF SCIENTIFIC COUNSELORS 10 00:00:32,115 --> 00:00:33,416 IS CALLED TO ORDER. 11 00:00:33,416 --> 00:00:35,418 AT MULTIPLE POINTS THE BOARD 12 00:00:35,418 --> 00:00:40,557 WILL GO INTO CLOSED SESSION. 13 00:00:40,557 --> 00:00:42,425 THE DESIGNATED STAFF WILL REMAIN 14 00:00:42,425 --> 00:00:44,728 IN THE MAIN TEAMS MEETING ROOM. 15 00:00:44,728 --> 00:00:45,462 EVERYBODY NOT DESIGNATED WILL 16 00:00:45,462 --> 00:00:48,665 NEED TO LOG OUT OF TEAMS UNTIL 17 00:00:48,665 --> 00:00:49,766 THE OPEN SESSION RESUMES AND 18 00:00:49,766 --> 00:00:51,268 DURING CLOSED SESSIONS I'LL ASK 19 00:00:51,268 --> 00:00:53,203 FOR VIDEOCAST TO BE DROPPED AND 20 00:00:53,203 --> 00:00:55,605 RECORDING TO BE STOPPED. 21 00:00:55,605 --> 00:00:58,975 SO, AGAIN, WELCOME, EVERYBODY. 22 00:00:58,975 --> 00:01:04,748 I'D LIKE TO WELCOME TOBIN 23 00:01:04,748 --> 00:01:06,383 TOESMICK, AND AD HOC REVIEWER 24 00:01:06,383 --> 00:01:08,018 TODAY. 25 00:01:08,018 --> 00:01:15,325 I WITH THAT I WILL TURN IT OVER 26 00:01:15,325 --> 00:01:17,360 TO -- STEVE SHERRY IS NOT HERE. 27 00:01:17,360 --> 00:01:19,729 DAVID, WERE YOU GOING TO COVER 28 00:01:19,729 --> 00:01:20,630 HIS REMARKS? 29 00:01:20,630 --> 00:01:22,332 >> RICHARD IS GOING TO SPEND 30 00:01:22,332 --> 00:01:24,768 THE NEXT SEVERAL MINUTES 31 00:01:24,768 --> 00:01:26,736 TALKING. 32 00:01:26,736 --> 00:01:27,838 >> GREAT. 33 00:01:27,838 --> 00:01:28,104 THANK YOU. 34 00:01:28,104 --> 00:01:28,772 >> ALL RIGHT. 35 00:01:28,772 --> 00:01:38,215 LET ME SHARE MY SCREEN. 36 00:01:38,215 --> 00:01:40,650 ARE YOU SEEING THE CORRECT 37 00:01:40,650 --> 00:01:40,884 VERSION? 38 00:01:40,884 --> 00:01:42,819 >> I'M SEEING A COVER SLIDE THAT 39 00:01:42,819 --> 00:01:46,756 SAYS NLM DIVISION OF INTRAMURAL 40 00:01:46,756 --> 00:01:47,691 RESEARCH RECENT ACCOMPLISHMENTS. 41 00:01:47,691 --> 00:01:49,025 >> OKAY, GREAT. 42 00:01:49,025 --> 00:01:49,826 ALL RIGHT. 43 00:01:49,826 --> 00:01:53,230 WELL, FIRST OF ALL, I WANT TO 44 00:01:53,230 --> 00:01:55,398 THANK EVERYBODY FOR ATTENDING. 45 00:01:55,398 --> 00:01:57,267 DR. SHERRY, OUR ACTING NLM 46 00:01:57,267 --> 00:02:00,136 DIRECTOR, IS NOT ABLE TO JOIN US 47 00:02:00,136 --> 00:02:03,607 TODAY BUT SENDS HIS REGARDS AND 48 00:02:03,607 --> 00:02:09,312 HE ALSO WANTED TO EXTEND THANKS 49 00:02:09,312 --> 00:02:10,947 TO THE BOARD MEMBERS FOR 50 00:02:10,947 --> 00:02:12,249 PARTICIPATING IN THIS BOARD. 51 00:02:12,249 --> 00:02:14,985 WE RELY ON YOUR CRITICAL 52 00:02:14,985 --> 00:02:15,952 EVALUATION OF THE RESEARCH 53 00:02:15,952 --> 00:02:19,589 PROGRAM THAT WE HAVE DEVELOPED 54 00:02:19,589 --> 00:02:20,790 HERE AT THE NATIONAL LIBRARY OF 55 00:02:20,790 --> 00:02:21,157 MEDICINE. 56 00:02:21,157 --> 00:02:23,460 IT'S IMPORTANT WE'RE ABLE TO 57 00:02:23,460 --> 00:02:24,394 DEMONSTRATE AN EFFECTIVE RETURN 58 00:02:24,394 --> 00:02:27,364 ON INVESTMENT THAT WE'RE MAKING 59 00:02:27,364 --> 00:02:29,099 IN BIOMEDICAL RESEARCH HERE AT 60 00:02:29,099 --> 00:02:31,001 THE NATIONAL LIBRARY, AND YOUR 61 00:02:31,001 --> 00:02:32,602 EVALUATION OF YOUR RESEARCH 62 00:02:32,602 --> 00:02:35,238 PROGRAMS IS REALLY CRITICAL FOR 63 00:02:35,238 --> 00:02:36,473 MAKING THAT DETERMINATION. 64 00:02:36,473 --> 00:02:40,677 SO, WE WANT TO THANK YOU FOR 65 00:02:40,677 --> 00:02:41,878 YOUR EFFORTS. 66 00:02:41,878 --> 00:02:44,748 SO LET'S SEE. 67 00:02:44,748 --> 00:02:45,215 NEXT SLIDE. 68 00:02:45,215 --> 00:02:50,553 SO, I ALSO WANT TO SPECIFICALLY 69 00:02:50,553 --> 00:02:54,991 THANK PETER TARCZY-HORNOCH, 70 00:02:54,991 --> 00:02:56,760 BONNIE BERGER, AND BRASIELA 71 00:02:56,760 --> 00:02:58,929 HERNANDEZ, THEY WILL BE LEAVING 72 00:02:58,929 --> 00:03:01,564 US AFTER OUR MEETING TODAY. 73 00:03:01,564 --> 00:03:03,867 WE APPRECIATE THE EFFORT YOU'VE 74 00:03:03,867 --> 00:03:05,001 PUT INTO PROVIDING THIS BOARD 75 00:03:05,001 --> 00:03:12,676 REVIEW FOR US IN THE PAST. 76 00:03:12,676 --> 00:03:16,179 IN ADDITION I THANK PETER FOR 77 00:03:16,179 --> 00:03:17,714 SERVING AS CHAIR FOR SEVERAL 78 00:03:17,714 --> 00:03:19,149 YEARS, HAS DONE AN OUTSTANDING 79 00:03:19,149 --> 00:03:22,452 JOB MAKING SURE WE'RE PROVIDING 80 00:03:22,452 --> 00:03:26,256 A FAIR AND COMPREHENSIVE 81 00:03:26,256 --> 00:03:31,928 EVALUATION SO WE APPRECIATE THE 82 00:03:31,928 --> 00:03:33,697 EFFORT OF PETER AND OTHER BOARD 83 00:03:33,697 --> 00:03:33,930 MEMBERS. 84 00:03:33,930 --> 00:03:35,765 THERE ARE THREE MEMBERS WHO WERE 85 00:03:35,765 --> 00:03:43,673 NOT -- WILL NOT BE JOINING US 86 00:03:43,673 --> 00:03:45,342 TODAY. 87 00:03:45,342 --> 00:03:48,745 NOEMIE ELHADAD, BEHNAZ GHORAANI 88 00:03:48,745 --> 00:03:50,046 AROUND JOAN-EMMA SHEA, THEY WERE 89 00:03:50,046 --> 00:03:51,314 TERMINATED BY THE NEW 90 00:03:51,314 --> 00:03:53,516 ADMINISTRATION FOR REASONS NOT 91 00:03:53,516 --> 00:03:54,517 CLEAR TO US. 92 00:03:54,517 --> 00:03:57,787 BUT THEY WILL NOTE BE JOINING US 93 00:03:57,787 --> 00:04:00,123 AT THE MEETING TODAY. 94 00:04:00,123 --> 00:04:01,858 I ALSO WANT TO -- AS A RESULT, 95 00:04:01,858 --> 00:04:03,994 SORRY, AS A RESULT OF THAT WE'RE 96 00:04:03,994 --> 00:04:07,097 ACTUALLY GOING TO BE -- WE'RE IN 97 00:04:07,097 --> 00:04:08,732 THE PROCESS OF RECRUITING SIX 98 00:04:08,732 --> 00:04:09,432 NEW BOARD MEMBERS. 99 00:04:09,432 --> 00:04:13,069 DAVID AND I HAVE BEEN WORKING 100 00:04:13,069 --> 00:04:15,438 WITH OUR INVESTIGATORS TO 101 00:04:15,438 --> 00:04:16,973 IDENTIFY APPROPRIATE CANDIDATES 102 00:04:16,973 --> 00:04:20,276 WHO HAVE THE EXPERTISE NECESSARY 103 00:04:20,276 --> 00:04:21,444 FOR EVALUATING OUR RESEARCH, AND 104 00:04:21,444 --> 00:04:23,013 WE'RE IN THE PROCESS OF REACHING 105 00:04:23,013 --> 00:04:24,447 OUT TO THEM AND RECRUITING THEM 106 00:04:24,447 --> 00:04:26,549 AND HOPEFULLY WE'LL HAVE THEM ON 107 00:04:26,549 --> 00:04:30,553 BOARD FOR OUR NEXT MEETING IN 108 00:04:30,553 --> 00:04:30,787 OCTOBER. 109 00:04:30,787 --> 00:04:36,760 AND FINALLY I WANTED TO ALSO 110 00:04:36,760 --> 00:04:37,394 WELCOME TOBIN SOSNICK, SERVING 111 00:04:37,394 --> 00:04:40,730 AS AD HOC MEMBER ON THE BOARD 112 00:04:40,730 --> 00:04:41,798 TODAY, HELPING WITH THE REVIEW 113 00:04:41,798 --> 00:04:44,501 OF ONE OF OUR TENURE TRACK 114 00:04:44,501 --> 00:04:47,070 INVESTIGATORS, LAUREN PORTER. 115 00:04:47,070 --> 00:04:49,906 SO THANKS AGAIN, EVERYONE, FOR 116 00:04:49,906 --> 00:04:54,878 YOUR ASSISTANCE IN THIS REVIEW. 117 00:04:54,878 --> 00:04:57,914 SO, I JUST WANTED TO USE THIS AS 118 00:04:57,914 --> 00:05:00,016 AN OPPORTUNITY TO GIVE YOU KIND 119 00:05:00,016 --> 00:05:03,686 OF A QUICK OVERVIEW OF THE 120 00:05:03,686 --> 00:05:06,623 OVERALL ACCOMPLISHMENTS OF OUR 121 00:05:06,623 --> 00:05:07,724 INTRAMURAL RESEARCH PROGRAM. 122 00:05:07,724 --> 00:05:12,028 THE DIVISION OF INTRAMURAL 123 00:05:12,028 --> 00:05:15,665 RESEARCH AT NLM HAS 17 124 00:05:15,665 --> 00:05:17,233 INVESTIGATORS, 10 SENIOR 125 00:05:17,233 --> 00:05:18,935 INVESTIGATORS, AND 7 TENURE 126 00:05:18,935 --> 00:05:19,736 TRACK INVESTIGATORS. 127 00:05:19,736 --> 00:05:30,113 THEIR PICTURES ARE ON THE SLIDE. 128 00:05:30,113 --> 00:05:30,513 SORRY. 129 00:05:30,513 --> 00:05:33,516 PROBLEMS WITH TEAMS. 130 00:05:33,516 --> 00:05:35,919 JUST FOR YOUR AWARENESSN 131 00:05:35,919 --> 00:05:38,655 FEBRUARY OF THIS YEAR TWO OF OUR 132 00:05:38,655 --> 00:05:43,159 TENURE TRACK INVESTIGATORS WERE 133 00:05:43,159 --> 00:05:46,863 PUT ON ADMINISTRATIVE LEAVE. 134 00:05:46,863 --> 00:05:49,499 FORTUNATELY WE WERE ABLE TO HAVE 135 00:05:49,499 --> 00:05:52,001 THAT -- THAT WAS PART OF THE 136 00:05:52,001 --> 00:05:52,969 PROBATIONARY TERMINATION THAT 137 00:05:52,969 --> 00:05:55,572 YOU MAY HAVE HEARD ABOUT THAT WE 138 00:05:55,572 --> 00:05:59,042 EXPERIENCED IN FEBRUARY AT NIH. 139 00:05:59,042 --> 00:06:00,944 FORTUNATELY WE WERE ABLE TO 140 00:06:00,944 --> 00:06:03,780 ARGUE FOR THE CRITICAL ROLE THAT 141 00:06:03,780 --> 00:06:06,916 OUR INVESTIGATORS ARE PLAYING IN 142 00:06:06,916 --> 00:06:10,653 NIH'S MISSION AND WE WERE ABLE 143 00:06:10,653 --> 00:06:11,588 TO GET BOTH INDIVIDUALS 144 00:06:11,588 --> 00:06:14,324 REINSTATED, AND SO WE HAVE OUR 145 00:06:14,324 --> 00:06:22,932 FULL COMPLEMENT OF INVESTIGATORS 146 00:06:22,932 --> 00:06:24,367 AS OF TODAY. 147 00:06:24,367 --> 00:06:25,368 SOMETHING ELSE THAT WE WANTED TO 148 00:06:25,368 --> 00:06:28,304 MAKE SURE THE BOARD WAS AWARE OF 149 00:06:28,304 --> 00:06:32,575 IS THAT WE HAVE HAD SOME 150 00:06:32,575 --> 00:06:33,276 ADJUSTMENTS TO OUR 151 00:06:33,276 --> 00:06:34,911 COMMUNICATIONS THAT WE'RE ABLE 152 00:06:34,911 --> 00:06:41,551 TO MAKE HERE AT NIH. 153 00:06:41,551 --> 00:06:44,120 IN JANUARY OF 24 A COMMUNICATION 154 00:06:44,120 --> 00:06:46,089 PAUSE WAS INSTITUTED AT NIH, 155 00:06:46,089 --> 00:06:51,728 THIS INCLUDED OUR ABILITY TO 156 00:06:51,728 --> 00:06:52,829 SUBMIT SCIENTIFIC MANUSCRIPTS 157 00:06:52,829 --> 00:06:54,497 FOR PEER-REVIEWED JOURNALS. 158 00:06:54,497 --> 00:06:55,698 THE COMMUNICATIONS HAVE SLOWLY 159 00:06:55,698 --> 00:07:00,737 BEEN OPENING UP SINCE THEN. 160 00:07:00,737 --> 00:07:02,972 FORTUNATELY, THE MORATORIUM ON 161 00:07:02,972 --> 00:07:03,606 MANUSCRIPT SUBMISSIONS WAS VERY 162 00:07:03,606 --> 00:07:05,175 BRIEF, IT WAS REALLY ONLY A 163 00:07:05,175 --> 00:07:06,442 MONTH OR TWO. 164 00:07:06,442 --> 00:07:08,578 AND SO WE HAVE NOW AFTER THAT 165 00:07:08,578 --> 00:07:13,249 BRIEF PAUSE BEEN ABLE TO SUBMIT 166 00:07:13,249 --> 00:07:13,816 MANUSCRIPTS FOR SCIENTIFIC 167 00:07:13,816 --> 00:07:16,252 PUBLICATION AND SO THAT HAS BEEN 168 00:07:16,252 --> 00:07:19,989 A VERY POSITIVE KIND OF RECENT 169 00:07:19,989 --> 00:07:20,223 CHANGE. 170 00:07:20,223 --> 00:07:22,926 KIND OF RELATED TO THE PAUSE IN 171 00:07:22,926 --> 00:07:26,095 COMMUNICATION WE WERE ALSO -- WE 172 00:07:26,095 --> 00:07:27,964 HAD A BAN ON TRAVEL TO 173 00:07:27,964 --> 00:07:29,332 SCIENTIFIC MEETINGS. 174 00:07:29,332 --> 00:07:31,634 AS OF APRIL 10 THAT BAN HAS BEEN 175 00:07:31,634 --> 00:07:33,803 LIFTED, SO NOW WE'RE ALLOWED TO 176 00:07:33,803 --> 00:07:37,373 ONCE AGAIN ENGAGE IN OFFICIAL 177 00:07:37,373 --> 00:07:40,743 TRAVEL, TO PARTICIPATE IN PUBLIC 178 00:07:40,743 --> 00:07:42,111 SCIENTIFIC CONFERENCES FOR 179 00:07:42,111 --> 00:07:43,446 PRESENTATION OF SCIENTIFIC DATA, 180 00:07:43,446 --> 00:07:47,183 OR FOR THE PURPOSES OF LEARNING 181 00:07:47,183 --> 00:07:50,420 ABOUT SCIENTIFIC RESEARCH. 182 00:07:50,420 --> 00:07:53,356 AND SO THOSE ACTIVITIES HAVE 183 00:07:53,356 --> 00:07:54,224 RESUMED. 184 00:07:54,224 --> 00:07:58,628 AND THEN RELATED TO THIS, WE 185 00:07:58,628 --> 00:08:01,497 ALSO HAVE -- WE'RE STILL KIND OF 186 00:08:01,497 --> 00:08:04,200 PAUSED IN OUR ABILITY TO ISSUE 187 00:08:04,200 --> 00:08:06,336 DOCUMENTS TO THE PUBLIC, 188 00:08:06,336 --> 00:08:08,271 ESPECIALLY DOCUMENTS THAT RELATE 189 00:08:08,271 --> 00:08:12,642 TO THINGS LIKE REGULATION, 190 00:08:12,642 --> 00:08:13,643 GUIDANCE, NOTICES, GRANT 191 00:08:13,643 --> 00:08:16,379 ANNOUNCEMENTS AND THINGS LIKE 192 00:08:16,379 --> 00:08:17,013 THAT. 193 00:08:17,013 --> 00:08:19,649 THOSE DOCUMENTS NEED TO BE 194 00:08:19,649 --> 00:08:21,050 SUBMITTED TO THE -- AT THE 195 00:08:21,050 --> 00:08:23,119 DEPARTMENT LEVEL FOR REVIEW AND 196 00:08:23,119 --> 00:08:25,755 APPROVAL BEFORE THEY CAN BE 197 00:08:25,755 --> 00:08:26,022 RELEASED. 198 00:08:26,022 --> 00:08:29,092 AND SO THERE IS KIND OF 199 00:08:29,092 --> 00:08:32,228 ADDITIONAL SCRUTINY ON PUBLIC 200 00:08:32,228 --> 00:08:33,863 COMMUNICATIONS THAT WE'RE 201 00:08:33,863 --> 00:08:34,998 CURRENTLY EXPERIENCING AND THAT 202 00:08:34,998 --> 00:08:37,066 JUST REQUIRES US TO GO THROUGH 203 00:08:37,066 --> 00:08:39,802 KIND OF THESE ADDITIONAL LEVELS 204 00:08:39,802 --> 00:08:42,405 OF REVIEW AND APPROVAL. 205 00:08:42,405 --> 00:08:43,306 THAT IS WORKING RELATIVELY 206 00:08:43,306 --> 00:08:44,073 SMOOTHLY BUT IT'S A SLIGHT 207 00:08:44,073 --> 00:08:48,778 CHANGE IN THE WAY WE'VE OPERATD 208 00:08:48,778 --> 00:08:50,280 IN THE PAST. 209 00:08:50,280 --> 00:08:53,082 NOW, IN SPITE OF THE KIND OF 210 00:08:53,082 --> 00:08:55,551 PAUSE IN COMMUNICATIONS, I THINK 211 00:08:55,551 --> 00:08:58,087 OUR INVESTIGATORS CONTINUE TO BE 212 00:08:58,087 --> 00:08:59,856 VERY PRODUCTIVE IN TERMS OF 213 00:08:59,856 --> 00:09:00,823 SCIENTIFIC PUBLICATIONS. 214 00:09:00,823 --> 00:09:05,061 SO SINCE THE BEGINNING OF 2025, 215 00:09:05,061 --> 00:09:06,763 WE'VE HAD 28 PEER-REVIEWED 216 00:09:06,763 --> 00:09:11,634 JOURNAL ARTICLES AND CONFERENCE 217 00:09:11,634 --> 00:09:13,436 PROCEEDINGS PUBLISHED, IN 218 00:09:13,436 --> 00:09:17,907 JOURNALS, HIGH-IMPACT JOURNALS 219 00:09:17,907 --> 00:09:20,543 LIKE NATURE, NATURE MEDICINE, 220 00:09:20,543 --> 00:09:23,112 CELL, MOLECULAR SCIENCE AND 221 00:09:23,112 --> 00:09:23,413 OTHERS. 222 00:09:23,413 --> 00:09:26,049 WE HAVE 53 SUBMISSIONS UNDER WE 223 00:09:26,049 --> 00:09:28,885 VIEW CURRENTLY, SO OUR 224 00:09:28,885 --> 00:09:30,320 INVESTIGATORS REMAIN HIGHLY 225 00:09:30,320 --> 00:09:34,223 PRODUCTIVE IN TERMS OF 226 00:09:34,223 --> 00:09:34,557 PUBLICATIONS. 227 00:09:34,557 --> 00:09:36,759 I'VE LISTED FOUR NOTEWORTHY 228 00:09:36,759 --> 00:09:39,329 EXAMPLES HERE ON THE SLIDE. 229 00:09:39,329 --> 00:09:42,065 LAUREN PORTER'S GROUP HAD A 230 00:09:42,065 --> 00:09:44,734 NATURE PAPER THIS YEAR WHERE 231 00:09:44,734 --> 00:09:48,371 THEY ARE REPORTING ON SOME 232 00:09:48,371 --> 00:09:53,743 LIMITATIONS TO THIS VERY POPULAR 233 00:09:53,743 --> 00:09:56,479 ALGORITHM CALLED ALPHAFOLD2 FOR 234 00:09:56,479 --> 00:09:58,548 PROTEIN STRUCTURE DETERMINATION, 235 00:09:58,548 --> 00:09:59,782 WHERE THEY FOUND CERTAIN ASPECTS 236 00:09:59,782 --> 00:10:01,884 OF THE ALGORITHM THAT INCLUDE 237 00:10:01,884 --> 00:10:03,986 SEQUENCE CLUSTERING HAVE AN 238 00:10:03,986 --> 00:10:05,455 IMPACT ON THE PERFORMANCE OF THE 239 00:10:05,455 --> 00:10:08,858 ALGORITHM AS YOU MIGHT EXPECT. 240 00:10:08,858 --> 00:10:16,632 WE ALSO HAVE A PUBLICATION BY 241 00:10:16,632 --> 00:10:18,301 PRITHCA, WHAT ROLE BIOLOGICAL 242 00:10:18,301 --> 00:10:20,470 DATABASES HAVE IN THE AGE OF 243 00:10:20,470 --> 00:10:22,071 GENERATIVE A.I. AND HOW 244 00:10:22,071 --> 00:10:23,406 IMPORTANT THESE KIND OF 245 00:10:23,406 --> 00:10:25,007 DATABASES REMAIN, IN THIS 246 00:10:25,007 --> 00:10:28,044 CURRENT YEAR OF ARTIFICIAL 247 00:10:28,044 --> 00:10:29,078 INTELLIGENCE. 248 00:10:29,078 --> 00:10:31,280 THERE WAS A PUBLICATION BY 249 00:10:31,280 --> 00:10:32,849 EUGENE KOONIN AND COLLABORATORS 250 00:10:32,849 --> 00:10:36,753 IN MOLECULAR CELL WHERE THEY ARE 251 00:10:36,753 --> 00:10:40,223 DEFINE A VERY INTERESTING NEW 252 00:10:40,223 --> 00:10:42,258 PHENOMENON, PRESENCE OF MICRO 253 00:10:42,258 --> 00:10:43,559 PROTEIN HIDDEN IN GENOMES, A NEW 254 00:10:43,559 --> 00:10:45,161 DISCOVERY THAT I THINK WILL HAVE 255 00:10:45,161 --> 00:10:47,130 AN IMPORTANT IMPACT ON OUR 256 00:10:47,130 --> 00:10:52,068 UNDERSTANDING OF THE ROLE OF 257 00:10:52,068 --> 00:10:54,670 PROTEOMES IN BACTERIAL 258 00:10:54,670 --> 00:10:54,971 PHYSIOLOGY. 259 00:10:54,971 --> 00:10:57,440 AND THEN FINALLY, FINAL EXAMPLE, 260 00:10:57,440 --> 00:10:59,876 IS A PUBLICATION IN ANNUAL 261 00:10:59,876 --> 00:11:01,911 REVIEWS OF BIOMEDICAL DATA 262 00:11:01,911 --> 00:11:06,549 SCIENCE BY JOHN LU AND TEAM, 263 00:11:06,549 --> 00:11:08,918 DEFINING THE CHALLENGES WITH 264 00:11:08,918 --> 00:11:10,987 USING AN IMPLEMENTING LARGE 265 00:11:10,987 --> 00:11:11,587 LANGUAGE MODELS IN HEALTHCARE 266 00:11:11,587 --> 00:11:14,991 SETTING, ONE OF THE KINDS -- 267 00:11:14,991 --> 00:11:16,893 WHAT ARE THE KINDS OF 268 00:11:16,893 --> 00:11:19,762 LIMITATIONS AND GUARD RAILS AS 269 00:11:19,762 --> 00:11:21,631 WE TRY TO ADOPT NEW METHODS. 270 00:11:21,631 --> 00:11:24,167 THESE ARE A COUPLE EXAMPLES OF 271 00:11:24,167 --> 00:11:27,537 PUBLICATIONS FROM THE DIVISION. 272 00:11:27,537 --> 00:11:28,671 MANY OF OUR INVESTIGATORS 273 00:11:28,671 --> 00:11:30,606 RECEIVED RECENT AWARDS AND 274 00:11:30,606 --> 00:11:32,575 HONORS, DAVID LANDSMAN, MY 275 00:11:32,575 --> 00:11:34,210 DEPUTY DIRECTOR, WAS ELECTED AS 276 00:11:34,210 --> 00:11:36,846 A FELLOW OF THE INTERNATIONAL 277 00:11:36,846 --> 00:11:37,980 SOCIETY OF COMPUTATIONAL BIOLOGY 278 00:11:37,980 --> 00:11:48,458 WHICH IS A VERY PRESTIGIOUS 279 00:11:49,292 --> 00:11:53,896 ORGANIZATION. 280 00:11:53,896 --> 00:11:55,064 DINA DEMNEH-FUSHMAN. 281 00:11:55,064 --> 00:12:02,905 LAUREN PORTER APPOINTED TO 282 00:12:02,905 --> 00:12:03,439 DIRECTOR. 283 00:12:03,439 --> 00:12:05,508 THERESA PRZYTYKKA ELECTED TO 284 00:12:05,508 --> 00:12:07,276 COLLEGE OF FELLOWS OF AMERICAN 285 00:12:07,276 --> 00:12:09,145 INSTITUTE FOR MEDICAL AND 286 00:12:09,145 --> 00:12:12,515 BIOLOGICAL ENGINEERING. 287 00:12:12,515 --> 00:12:23,025 EUGENE KOONIN AWARDED HONORARY 288 00:12:23,292 --> 00:12:24,227 DOCTORATES. 289 00:12:24,227 --> 00:12:26,162 AUGUSTINE LUNA SELECTED AS 290 00:12:26,162 --> 00:12:28,698 ORGANIZER FOR ISMB 2025 MEETING 291 00:12:28,698 --> 00:12:30,032 AND COMMUNITY COORDINATOR FOR 292 00:12:30,032 --> 00:12:32,235 THE COMBINED INITIATIVE WHICH IS 293 00:12:32,235 --> 00:12:34,937 COORDINATING DEVELOPMENT OF 294 00:12:34,937 --> 00:12:36,172 COMMUNITY STANDARDS AROUND 295 00:12:36,172 --> 00:12:36,639 COMPUTATIONAL MODELS. 296 00:12:36,639 --> 00:12:38,975 FINALLY I WAS PROMOTED TO SENIOR 297 00:12:38,975 --> 00:12:43,145 INVESTIGATOR WITH TENURE AT NIH 298 00:12:43,145 --> 00:12:44,714 EARLIER THIS YEAR. 299 00:12:44,714 --> 00:12:47,250 OUR TRAININGS HAVE BEEN QUITE 300 00:12:47,250 --> 00:12:50,186 SUCCESSFUL, AND SO THESE ARE NOW 301 00:12:50,186 --> 00:12:52,855 TRAINEES GETTING READY TO EMBARK 302 00:12:52,855 --> 00:12:56,626 ON THE NEXT PHASE OF THEIR 303 00:12:56,626 --> 00:12:59,362 SCIENTIFIC CAREER, ONE 304 00:12:59,362 --> 00:13:04,901 POSTDOCTORAL FELLOW OLIVER BEAR 305 00:13:04,901 --> 00:13:06,569 DON'T WALK OFFERED ASSISTANT 306 00:13:06,569 --> 00:13:09,572 PROFESSOR POSITION IN DEPARTMENT 307 00:13:09,572 --> 00:13:11,240 OF BIOMEDICAL MEDICAL 308 00:13:11,240 --> 00:13:11,774 INFORMATICS UNIVERSITY OF 309 00:13:11,774 --> 00:13:13,976 WASHINGTON, THAT WAS A GREAT 310 00:13:13,976 --> 00:13:15,511 ACCOMPLISHMENT. 311 00:13:15,511 --> 00:13:26,055 AND THEN I'VE LISTED A NUMBER OF 312 00:13:26,489 --> 00:13:28,124 POSTBAC FELLOWS IN A NUMBER OF 313 00:13:28,124 --> 00:13:30,226 UNIVERSITIES, I WON'T GO THROUGH 314 00:13:30,226 --> 00:13:35,131 ALL IN DETAIL BUT JUST TO 315 00:13:35,131 --> 00:13:38,134 MENTION THE FACT THAT OUR 316 00:13:38,134 --> 00:13:41,237 POSTBAC TRAINING PROGRAM HAS 317 00:13:41,237 --> 00:13:42,471 ALSO BEEN VERY SUCCESSFUL. 318 00:13:42,471 --> 00:13:45,408 SO, THE OTHER THING THAT I 319 00:13:45,408 --> 00:13:50,446 WANTED TO COVER BRIEFLY IS OUR 320 00:13:50,446 --> 00:13:52,815 NLM DIR ROADMAP. 321 00:13:52,815 --> 00:13:54,750 I'VE TALKED ABOUT THIS IN PAST 322 00:13:54,750 --> 00:13:56,752 MEETINGS BUT I'M HAPPY TO REPORT 323 00:13:56,752 --> 00:13:59,055 THAT THE ROADMAP HAS ACTUALLY 324 00:13:59,055 --> 00:14:02,625 BEEN COMPLETED TODAY, AS OF 325 00:14:02,625 --> 00:14:03,993 TODAY, APRIL 30. 326 00:14:03,993 --> 00:14:10,766 WE HAVE PRODUCED TWO VERSIONS OF 327 00:14:10,766 --> 00:14:11,167 THE ROADMAP. 328 00:14:11,167 --> 00:14:13,402 ONE THAT INCLUDES THE VISION AND 329 00:14:13,402 --> 00:14:15,671 MISSION OF THE DIR AS WELL AS A 330 00:14:15,671 --> 00:14:19,442 NUMBER OF GOALS AND SPECIFIC 331 00:14:19,442 --> 00:14:19,842 OBJECTIVES. 332 00:14:19,842 --> 00:14:22,478 THIS VERSION OF THAT DOCUMENT 333 00:14:22,478 --> 00:14:24,347 WILL BE MADE AVAILABLE FOR 334 00:14:24,347 --> 00:14:25,448 DISTRIBUTION TO THE PUBLIC. 335 00:14:25,448 --> 00:14:30,586 AND THEN WE HAVE A SECOND 336 00:14:30,586 --> 00:14:32,455 VERSION THAT IN ADDITION TO THE 337 00:14:32,455 --> 00:14:33,856 MISSION GOALS AND OBJECTIVES 338 00:14:33,856 --> 00:14:36,025 INCLUDES SPECIFIC IMPLEMENTATION 339 00:14:36,025 --> 00:14:36,959 PLANS AND ASSESSMENT METRICS 340 00:14:36,959 --> 00:14:40,229 THAT WE WILL BE USING FOR 341 00:14:40,229 --> 00:14:41,664 INTERNAL USE THAT PROVIDES SOME 342 00:14:41,664 --> 00:14:44,033 OF THE DETAILS ABOUT HOW WE'RE 343 00:14:44,033 --> 00:14:46,102 ACTUALLY GOING TO ACCOMPLISH 344 00:14:46,102 --> 00:14:48,771 THESE GOALS AND OBJECTIVES. 345 00:14:48,771 --> 00:14:53,542 SO, I JUST WANTED TO LIST HERE 346 00:14:53,542 --> 00:14:55,878 VISION AND MISSION STATEMENTS, 347 00:14:55,878 --> 00:14:58,147 VISION TO GROW AS A CENTER OF 348 00:14:58,147 --> 00:15:01,083 EXCELLENCE FOR INNOVATION AND 349 00:15:01,083 --> 00:15:03,753 COMPUTATIONAL BIOLOGY AND HELP 350 00:15:03,753 --> 00:15:04,720 INFORMATICS RESEARCH FOR 351 00:15:04,720 --> 00:15:06,489 INVESTIGATORS ENGAGED IN A 352 00:15:06,489 --> 00:15:07,623 VIBRANT, DIVERSE, COLLABORATIVE 353 00:15:07,623 --> 00:15:09,892 COMMUNITY TO ADVANCE BIOMEDICAL 354 00:15:09,892 --> 00:15:12,595 KNOWLEDGE AND IMPROVE HUMAN 355 00:15:12,595 --> 00:15:12,828 HEALTH. 356 00:15:12,828 --> 00:15:15,998 AND WE ARE ACCOMPLISHING THAT 357 00:15:15,998 --> 00:15:17,533 VISION IN A MANNER CONSISTENT 358 00:15:17,533 --> 00:15:23,272 WITH OUR MISSION, AND THAT IS TO 359 00:15:23,272 --> 00:15:26,108 SERVE AS - TO RESEARCH AND 360 00:15:26,108 --> 00:15:26,676 DEVELOP COMPUTATIONAL 361 00:15:26,676 --> 00:15:28,644 DEVELOPMENT TO A BROAD RANGE OF 362 00:15:28,644 --> 00:15:32,214 INFORMATION AND DATA SCIENCE 363 00:15:32,214 --> 00:15:32,682 PROBLEMS IN BIOLOGY, 364 00:15:32,682 --> 00:15:35,051 BIOMEDICINE, HUMAN HEALTH. 365 00:15:35,051 --> 00:15:36,585 AND LEVERAGING UNIQUE REVERSES 366 00:15:36,585 --> 00:15:38,220 AT THE NATIONAL LIBRARY OF 367 00:15:38,220 --> 00:15:40,823 MEDICINE AS A KEY ASPECT OF OUR 368 00:15:40,823 --> 00:15:42,058 RESEARCH PROGRAM. 369 00:15:42,058 --> 00:15:44,593 AS PART OF THAT MISSION WE 370 00:15:44,593 --> 00:15:46,162 PROMOTE THE FIELDS OF 371 00:15:46,162 --> 00:15:48,731 COMPUTATIONAL BIOLOGY AND HEALTH 372 00:15:48,731 --> 00:15:50,599 INFORMATICS THROUGH VARIOUS 373 00:15:50,599 --> 00:15:51,901 OUTREACH AND TRAINING ACTIVITIES 374 00:15:51,901 --> 00:15:53,869 THAT FOSTER AN ENVIRONMENT THAT 375 00:15:53,869 --> 00:15:56,439 ATTRACTS THE DIVERSE TALENT AND 376 00:15:56,439 --> 00:15:59,108 PERSPECTIVE NEEDED TO TACKLE THE 377 00:15:59,108 --> 00:16:00,342 CHALLENGING BIOMEDICAL PROBLEMS. 378 00:16:00,342 --> 00:16:02,778 SO THESE ARE KIND OF OUR VISION 379 00:16:02,778 --> 00:16:04,847 AND MISSION STATEMENTS THAT 380 00:16:04,847 --> 00:16:09,118 SERVE AS KIND OF THE NORTH STAR 381 00:16:09,118 --> 00:16:10,586 FOR OUR ROADMAP. 382 00:16:10,586 --> 00:16:13,656 AND THEN WE HAVE IDENTIFIED AS 383 00:16:13,656 --> 00:16:15,257 PART OF THE ROADMAP FIVE HIGH 384 00:16:15,257 --> 00:16:17,827 LEVEL GOALS IN A SERIES OF 385 00:16:17,827 --> 00:16:19,195 SPECIFIC OBJECTIVES THAT WILL 386 00:16:19,195 --> 00:16:23,365 SERVE AS KIND OF OUR BLUEPRINT. 387 00:16:23,365 --> 00:16:28,170 WE WANT TO GROW THE DIR 388 00:16:28,170 --> 00:16:29,572 COMPUTATIONAL RESEARCH BY 389 00:16:29,572 --> 00:16:31,440 IDENTIFYING AND REALIGNING 390 00:16:31,440 --> 00:16:32,641 RESEARCH PORTFOLIO TO HIGH 391 00:16:32,641 --> 00:16:34,610 PRIORITY RESEARCH AREAS, TO KIND 392 00:16:34,610 --> 00:16:37,480 OF FILL ANY GAPS THAT WE MIGHT 393 00:16:37,480 --> 00:16:41,016 HAVE IN KIND OF EMERGING 394 00:16:41,016 --> 00:16:43,619 COMPUTATIONAL RESEARCH AREAS. 395 00:16:43,619 --> 00:16:46,822 AND ALSO TO EMPHASIZE CROSS-NIH 396 00:16:46,822 --> 00:16:49,325 COLLABORATIONS AS WELL AS HIGHLY 397 00:16:49,325 --> 00:16:51,060 INNOVATIVE HIGH RISK AND HIGH 398 00:16:51,060 --> 00:16:54,597 REWARD TYPE PROJECTS THAT I 399 00:16:54,597 --> 00:16:58,434 THINK WE'RE UNIQUELY POSITIONED 400 00:16:58,434 --> 00:16:58,868 TO TACKLE. 401 00:16:58,868 --> 00:17:03,339 SECOND GOAL IS TO ADVANCE 402 00:17:03,339 --> 00:17:05,407 INNOVATIVE RESEARCH AND 403 00:17:05,407 --> 00:17:07,977 DEVELOPMENT PROJECTS, TO 404 00:17:07,977 --> 00:17:08,744 COLLABORATE WITH ENGINEERING 405 00:17:08,744 --> 00:17:12,648 BRANCHES, BOTH AT THE NCBI AND 406 00:17:12,648 --> 00:17:15,751 LISTER HILL ENGINEERING 407 00:17:15,751 --> 00:17:18,988 DIVISIONS, TO IDENTIFY KIND OF 408 00:17:18,988 --> 00:17:21,157 RESEARCH PRODUCTS THAT HAVE BEEN 409 00:17:21,157 --> 00:17:23,492 VALIDATED TO THE EXTENT THAT 410 00:17:23,492 --> 00:17:26,195 THEY COULD BE USEFUL PRODUCTION 411 00:17:26,195 --> 00:17:27,897 LEVEL RESOURCES THAT WE COULD 412 00:17:27,897 --> 00:17:31,801 RELEASE TO THE PUBLIC THROUGH 413 00:17:31,801 --> 00:17:33,969 OUR ENGINEERING BRANCHES AND 414 00:17:33,969 --> 00:17:34,703 THEIR PRODUCTION SYSTEMS. 415 00:17:34,703 --> 00:17:37,373 AND SO THAT IS REALLY TO TRY TO 416 00:17:37,373 --> 00:17:40,810 EMPHASIZE THE TRANSLATION OF OUR 417 00:17:40,810 --> 00:17:43,279 RESEARCH ACTIVITIES INTO 418 00:17:43,279 --> 00:17:44,513 PRODUCTION DEVELOPMENT. 419 00:17:44,513 --> 00:17:46,182 THE THIRD HIGH LEVEL GOAL IS 420 00:17:46,182 --> 00:17:50,119 TRAINING THE NEXT GENERATION OF 421 00:17:50,119 --> 00:17:51,587 COMPUTATIONAL SCIENTISTS TO NOT 422 00:17:51,587 --> 00:17:53,823 ONLY TRAIN THEM IN THE 423 00:17:53,823 --> 00:17:55,591 COMPUTATIONAL METHODS BUT ALSO 424 00:17:55,591 --> 00:17:58,294 HOW TO CONDUCT REPRODUCIBLE 425 00:17:58,294 --> 00:18:00,229 SCIENCE AND TO HELP THEM DEVELOP 426 00:18:00,229 --> 00:18:04,233 THE SKILLS THAT ARE NECESSARY 427 00:18:04,233 --> 00:18:05,167 FOR SUBSEQUENT CAREER 428 00:18:05,167 --> 00:18:08,337 ADVANCEMENT IN TERMS OF 429 00:18:08,337 --> 00:18:08,938 MANUSCRIPT PREPARATION, GRANT 430 00:18:08,938 --> 00:18:10,940 PROPOSAL DEVELOPMENT, AND THOSE 431 00:18:10,940 --> 00:18:14,109 KINDS OF -- KIND OF RUNNING A 432 00:18:14,109 --> 00:18:15,978 RESEARCH GROUP, THOSE KINDS OF 433 00:18:15,978 --> 00:18:16,645 ANCILLARY SKILLS THAT ARE 434 00:18:16,645 --> 00:18:20,883 NECESSARY TO BE A PRODUCTIVE 435 00:18:20,883 --> 00:18:21,150 SCIENTIST. 436 00:18:21,150 --> 00:18:23,752 FOURTH HIGH LEVEL GOAL TO RAISE 437 00:18:23,752 --> 00:18:25,354 THE VISIBILITY OF DIR RESEARCH 438 00:18:25,354 --> 00:18:27,456 SO THE COMMUNITY IS AWARE OF THE 439 00:18:27,456 --> 00:18:28,991 KINDS OF INNOVATIVE WORK WE DO 440 00:18:28,991 --> 00:18:34,263 HERE AT THE NATIONAL LIBRARY OF 441 00:18:34,263 --> 00:18:34,864 MEDICINE. 442 00:18:34,864 --> 00:18:38,667 FINALLY TO MODERNIZE OUR DIR 443 00:18:38,667 --> 00:18:40,069 ADMINISTRATION, ESPECIALLY IN 444 00:18:40,069 --> 00:18:42,404 THE CURRENT ENVIRONMENT, WHERE 445 00:18:42,404 --> 00:18:44,573 THERE'S PRETTY SIGNIFICANT 446 00:18:44,573 --> 00:18:45,574 REORGANIZATION OF ADMINISTRATION 447 00:18:45,574 --> 00:18:45,808 AT NIH. 448 00:18:45,808 --> 00:18:48,077 AND SO WE'RE GOING TO NEED TO BE 449 00:18:48,077 --> 00:18:52,014 ABLE TO ADAPT TO THOSE CHANGES 450 00:18:52,014 --> 00:18:53,515 THAT WE'RE EXPERIENCING HERE. 451 00:18:53,515 --> 00:18:56,085 SO THOSE ARE KIND OF THE FIVE 452 00:18:56,085 --> 00:18:58,354 HIGH LEVEL GOALS, AND SOME OF 453 00:18:58,354 --> 00:19:00,723 THE SPECIFIC OBJECTIVES THAT ARE 454 00:19:00,723 --> 00:19:03,058 GOING TO SERVE AS, YOU KNOW, OUR 455 00:19:03,058 --> 00:19:04,059 BLUEPRINT FOR MOVING FORWARD AND 456 00:19:04,059 --> 00:19:06,896 I HAVE TO SAY I THINK WE'RE VERY 457 00:19:06,896 --> 00:19:09,565 EXCITED ABOUT KIND OF 458 00:19:09,565 --> 00:19:12,201 REEMBARKING ON THIS ACTIVITY AND 459 00:19:12,201 --> 00:19:16,238 DRIVING OUR DEVELOPMENT MOVING 460 00:19:16,238 --> 00:19:16,772 FORWARD. 461 00:19:16,772 --> 00:19:19,775 SO, JUST A QUICK SUMMARY. 462 00:19:19,775 --> 00:19:21,277 HERE'S THE AGENDA THAT ALL OF 463 00:19:21,277 --> 00:19:23,345 YOU RECEIVED IN ADVANCE. 464 00:19:23,345 --> 00:19:26,682 WE'RE GOING TO BE REVIEWING ONE 465 00:19:26,682 --> 00:19:28,317 OF OUR SENIOR INVESTIGATORS, 466 00:19:28,317 --> 00:19:31,487 CLEM MCDONALD, AND TWO OF OUR 467 00:19:31,487 --> 00:19:32,254 JUNIOR TENURE TRACK 468 00:19:32,254 --> 00:19:33,555 INVESTIGATORS, JEREMY WEISS AND 469 00:19:33,555 --> 00:19:35,324 LAUREN PORTER. 470 00:19:35,324 --> 00:19:41,697 AND YOU CAN SEE THE SCHEDULE ON 471 00:19:41,697 --> 00:19:42,631 THE AGENDA HERE. 472 00:19:42,631 --> 00:19:44,533 THE OTHER THING THAT WE WANTED 473 00:19:44,533 --> 00:19:48,737 TO JUST PUT ON THE TABLE IS THAT 474 00:19:48,737 --> 00:19:51,106 FOR OUR UPCOMING OCTOBER 2025 475 00:19:51,106 --> 00:19:52,942 MEETING WE ARE PLANNING ON 476 00:19:52,942 --> 00:19:55,577 HAVING FOUR OF OUR INVESTIGATORS 477 00:19:55,577 --> 00:19:57,680 REVIEWED AND THE PLAN IS TO HAVE 478 00:19:57,680 --> 00:20:01,951 REVIEW CONDUCTED IN PERSON AS WE 479 00:20:01,951 --> 00:20:05,454 HAVE DONE HISTORICALLY, YOU 480 00:20:05,454 --> 00:20:08,090 KNOW, PRIOR TO COVID. 481 00:20:08,090 --> 00:20:09,558 BUT BECAUSE WE HAVE FOUR 482 00:20:09,558 --> 00:20:10,359 INVESTIGATORS AND WE'RE PLANNING 483 00:20:10,359 --> 00:20:14,863 ON DOING THIS IN PERSON, WE'RE 484 00:20:14,863 --> 00:20:17,700 PLANNING ON HAVING THE MEETING 485 00:20:17,700 --> 00:20:19,435 LAST FOR A DAY AND A HALF. 486 00:20:19,435 --> 00:20:22,104 AND SO THERE ARE TWO OPTIONS 487 00:20:22,104 --> 00:20:22,438 HERE. 488 00:20:22,438 --> 00:20:25,207 ONE IS WEDNESDAY AND THURSDAY 489 00:20:25,207 --> 00:20:28,477 OCTOBER 15 AND 16, OR THURSDAY 490 00:20:28,477 --> 00:20:30,079 AND FRIDAY, OCTOBER 16 AND 17. 491 00:20:30,079 --> 00:20:33,949 AND WE'LL SEND OUT A POLL TO GET 492 00:20:33,949 --> 00:20:36,552 FEEDBACK FROM THE BOARD ABOUT 493 00:20:36,552 --> 00:20:42,891 WHICH WOULD BE YOUR PREFERRED 494 00:20:42,891 --> 00:20:43,993 TIME FRAME. 495 00:20:43,993 --> 00:20:54,470 TWO OF THESE FOUR ARE TENURE 496 00:20:57,740 --> 00:20:57,873 TRACK. 497 00:20:57,873 --> 00:21:02,878 SHANG AND LUNA ARE TWO OF OUR 498 00:21:02,878 --> 00:21:04,680 TENURE TRACK INVESTIGATORS. 499 00:21:04,680 --> 00:21:06,348 YOU'VE RECEIVED INSTRUCTIONS 500 00:21:06,348 --> 00:21:10,686 ABOUT APPROACH TO REVIEWS, 501 00:21:10,686 --> 00:21:14,223 INCLUDING THE SCORING RANGES AND 502 00:21:14,223 --> 00:21:15,057 REVIEW CRITERIA WHICH INCLUDE 503 00:21:15,057 --> 00:21:18,961 EVALUATING THE QUALITY AND 504 00:21:18,961 --> 00:21:22,197 IMPACT OF OUR INVESTIGATOR 505 00:21:22,197 --> 00:21:23,832 SCIENCE, EVALUATING THEIR 506 00:21:23,832 --> 00:21:26,869 INTELLECTUAL LEADERSHIP IN THE 507 00:21:26,869 --> 00:21:29,104 FIELD, PUBLICATION RECORD, 508 00:21:29,104 --> 00:21:29,905 PROFESSIONAL LEADERSHIP, AND 509 00:21:29,905 --> 00:21:31,473 AWARDS AND HONORS THEY HAVE 510 00:21:31,473 --> 00:21:36,945 RECEIVED, AND THEIR TRAINING AND 511 00:21:36,945 --> 00:21:37,813 MENTORING ACTIVITIES. 512 00:21:37,813 --> 00:21:40,382 WE DID WANT TO REMIND THE BOARD 513 00:21:40,382 --> 00:21:43,585 THAT WE WANT EACH OF OUR 514 00:21:43,585 --> 00:21:45,487 INVESTIGATORS TO BE EVALUATED ON 515 00:21:45,487 --> 00:21:50,125 THEIR OWN MERITS. 516 00:21:50,125 --> 00:21:55,064 AND THAT MEANS AVOID COMPARING 517 00:21:55,064 --> 00:21:55,564 ACCOMPLISHMENTS BETWEEN 518 00:21:55,564 --> 00:21:56,498 INVESTIGATORS AT THE DIVISION. 519 00:21:56,498 --> 00:22:00,069 WE WANT EACH INDIVIDUAL TO BE 520 00:22:00,069 --> 00:22:02,171 EVALUATED ON THEIR OWN 521 00:22:02,171 --> 00:22:02,671 ACCOMPLISHMENTS. 522 00:22:02,671 --> 00:22:06,375 AND SO THAT'S AN IMPORTANT 523 00:22:06,375 --> 00:22:08,210 CONSIDERATION. 524 00:22:08,210 --> 00:22:11,513 ESPECIALLY FOR OUR TENURE TRACK 525 00:22:11,513 --> 00:22:12,981 INVESTIGATORS. 526 00:22:12,981 --> 00:22:14,716 FINALLY, I DID SUMMARIZE SOME 527 00:22:14,716 --> 00:22:17,686 RECENT CHALLENGES WE'VE BEEN 528 00:22:17,686 --> 00:22:19,421 NAVIGATING AROUND HIRING 529 00:22:19,421 --> 00:22:21,156 FREEZES, COMMUNICATION FREEZES, 530 00:22:21,156 --> 00:22:23,325 AND OTHER CONSIDERATIONS THAT 531 00:22:23,325 --> 00:22:26,161 COULD POTENTIALLY HAVE AN IMPACT 532 00:22:26,161 --> 00:22:28,764 ON THE ACTIVITIES OF OUR 533 00:22:28,764 --> 00:22:29,498 INVESTIGATORS, ESPECIALLY THE 534 00:22:29,498 --> 00:22:31,800 START-UP OF SOME OF OUR TENURE 535 00:22:31,800 --> 00:22:33,669 TRACK INVESTIGATORS, SO WE 536 00:22:33,669 --> 00:22:35,871 WANTED TO ASK THE BOARD TO, YOU 537 00:22:35,871 --> 00:22:37,372 KNOW, KEEP IN MIND IMPACT THAT 538 00:22:37,372 --> 00:22:42,678 SOME OF THOSE THINGS MAY HAVE 539 00:22:42,678 --> 00:22:47,116 HAD ON OUR RESEARCH 540 00:22:47,116 --> 00:22:47,483 ACCOMPLISHMENTS. 541 00:22:47,483 --> 00:22:48,750 SO, WITH THAT, IF THERE'S SOME 542 00:22:48,750 --> 00:22:50,519 MORE TIME, I'M HAPPY TO ANSWER 543 00:22:50,519 --> 00:22:52,621 ANY QUESTIONS THAT THE BOARD 544 00:22:52,621 --> 00:22:55,657 MIGHT HAVE BUT, AGAIN, I WANT TO 545 00:22:55,657 --> 00:22:56,425 THANK EVERYBODY FOR VOLUNTEERING 546 00:22:56,425 --> 00:22:58,026 THEIR TIME AND EFFORT TO HELPING 547 00:22:58,026 --> 00:23:02,698 US OUT WITH THESE EVALUATIONS. 548 00:23:02,698 --> 00:23:10,572 SO THANK YOU. 549 00:23:10,572 --> 00:23:11,773 >> A HAND IS UP. 550 00:23:11,773 --> 00:23:13,442 >> THANK YOU SO MUCH FOR THE 551 00:23:13,442 --> 00:23:13,675 SUMMARY. 552 00:23:13,675 --> 00:23:17,446 ALWAYS GOOD TO HEAR FROM YOU 553 00:23:17,446 --> 00:23:19,448 GUYS, AMAZING ACTIVITY AND 554 00:23:19,448 --> 00:23:20,883 REMARKABLE -- 555 00:23:20,883 --> 00:23:22,518 >> CAN YOU HOLD ON A SECOND. 556 00:23:22,518 --> 00:23:23,285 WE'RE HAVING TROUBLE HEARING 557 00:23:23,285 --> 00:23:23,552 YOU. 558 00:23:23,552 --> 00:23:29,191 I'M GOING TO TURN THE VOLUME UP. 559 00:23:29,191 --> 00:23:29,858 OKAY, TRY AGAIN. 560 00:23:29,858 --> 00:23:33,128 >> THANK YOU FOR THE SUMMARY. 561 00:23:33,128 --> 00:23:34,463 IMPRESSIVE RESEARCH, AS ALWAYS 562 00:23:34,463 --> 00:23:35,297 AT NLM, NOT SURPRISING. 563 00:23:35,297 --> 00:23:37,699 THANK YOU SO MUCH FOR THAT. 564 00:23:37,699 --> 00:23:39,535 I HAVE A QUICK QUESTION AND I 565 00:23:39,535 --> 00:23:41,036 COULD ASK PETER BUT I ALSO THINK 566 00:23:41,036 --> 00:23:43,872 I WANTED TO ASK DAVID THIS 567 00:23:43,872 --> 00:23:44,540 QUESTION, JUST BECAUSE IF I 568 00:23:44,540 --> 00:23:45,374 DON'T KNOW HE'S GOING TO BE 569 00:23:45,374 --> 00:23:48,043 LATER IN THE ROOM. 570 00:23:48,043 --> 00:23:50,479 FOR TWO TENURE TRACK 571 00:23:50,479 --> 00:23:52,548 INVESTIGATORS DO WE HAVE TO USE 572 00:23:52,548 --> 00:23:53,515 SPECIFIC LANGUAGE ABOUT THEIR 573 00:23:53,515 --> 00:23:54,616 PROMOTION TO TENURE OR ANYTHING 574 00:23:54,616 --> 00:23:55,684 LIKE THAT? 575 00:23:55,684 --> 00:23:58,320 ARE WE REQUIRED TO ALSO LOOK AT 576 00:23:58,320 --> 00:23:59,388 THEIR FILES IN THAT CONTEXT? 577 00:23:59,388 --> 00:24:04,259 AND DO YOU WANT US TO HAVE A 578 00:24:04,259 --> 00:24:06,195 SESSION IN OUR -- SPECIFICALLY 579 00:24:06,195 --> 00:24:07,429 IN OUR REPORT ABOUT THAT 580 00:24:07,429 --> 00:24:07,963 PARTICULAR ITEM? 581 00:24:07,963 --> 00:24:10,632 BECAUSE IN THE PAST WE HAVE HAD 582 00:24:10,632 --> 00:24:12,734 THAT ISSUE WHERE WE DIDN'T KNOW 583 00:24:12,734 --> 00:24:15,337 ABOUT IT UNTIL THE END, AND THEN 584 00:24:15,337 --> 00:24:17,806 WE WEREN'T SURE, YOU KNOW, WE 585 00:24:17,806 --> 00:24:19,441 REALLY DIDN'T SEE THE CANDIDATES 586 00:24:19,441 --> 00:24:20,142 IN THAT LIGHT. 587 00:24:20,142 --> 00:24:23,345 I THINK IT'S GOOD TO BRING IT UP 588 00:24:23,345 --> 00:24:24,179 RIGHT NOW. 589 00:24:24,179 --> 00:24:24,780 THANK YOU. 590 00:24:24,780 --> 00:24:26,481 >> OKAY, I CAN ANSWER THAT 591 00:24:26,481 --> 00:24:27,583 QUESTION. 592 00:24:27,583 --> 00:24:29,618 SO, TWO OF OUR INVESTIGATORS 593 00:24:29,618 --> 00:24:32,221 TODAY ARE TENURE TRACK, LAUREN 594 00:24:32,221 --> 00:24:34,323 PORTER AND JEREMY WEISS. 595 00:24:34,323 --> 00:24:37,092 JEREMY IS -- THIS IS HIS FIRST 596 00:24:37,092 --> 00:24:40,729 REVIEW, SO HE WILL BE -- HE'S 597 00:24:40,729 --> 00:24:44,099 ABOUT TWO YEARS INTO THE 598 00:24:44,099 --> 00:24:47,369 PROGRAM, SO THE STATEMENT 599 00:24:47,369 --> 00:24:50,706 PARAGRAPH OR SOMETHING LIKE THAT 600 00:24:50,706 --> 00:24:52,741 SAYING -- MENTIONING HIS 601 00:24:52,741 --> 00:24:55,310 PROGRESS IN THE FIRST TWO YEARS 602 00:24:55,310 --> 00:24:57,646 WHICH IS NORMALLY A LAG TIME AND 603 00:24:57,646 --> 00:25:00,549 THEN MAKE A STATEMENT ON TRACK 604 00:25:00,549 --> 00:25:05,420 OR NOT ON TRACK FOR TENURE 605 00:25:05,420 --> 00:25:05,754 CONSIDERATION. 606 00:25:05,754 --> 00:25:07,556 WHEREAS WITH LAUREN, SHE'S CLOSE 607 00:25:07,556 --> 00:25:10,826 TO SIX YEARS, IF NOT ALREADY SIX 608 00:25:10,826 --> 00:25:12,194 YEARS IN THE PROGRAM. 609 00:25:12,194 --> 00:25:17,366 AND SO WE'D LIKE SOME SHORT 610 00:25:17,366 --> 00:25:20,135 PARAGRAPH SUPPORTING THE BSC 611 00:25:20,135 --> 00:25:22,671 SUPPORT FOR TENURE TO PROCEED ON 612 00:25:22,671 --> 00:25:23,905 TO THE TENURE, TO RECOMMEND HER 613 00:25:23,905 --> 00:25:26,975 FOR TENURE. 614 00:25:26,975 --> 00:25:31,313 THIS IS THE FIRST PART OF THE 615 00:25:31,313 --> 00:25:32,547 TENURE REVIEW PROCESS. 616 00:25:32,547 --> 00:25:35,017 THE BOARD SAYS YES, THIS PERSON 617 00:25:35,017 --> 00:25:37,085 IS READY, THEN WE GO THROUGH 618 00:25:37,085 --> 00:25:39,221 NINE MONTHS OF PAPERWORK TO GO 619 00:25:39,221 --> 00:25:43,458 THROUGH THE TENURE PROCESS. 620 00:25:43,458 --> 00:25:45,560 BUT YES, WE'D APPRECIATE 621 00:25:45,560 --> 00:25:48,730 HAVING -- FOR EACH OF THOSE TWO 622 00:25:48,730 --> 00:25:53,035 PEOPLE, A PARAGRAPH OR 623 00:25:53,035 --> 00:25:54,870 STATEMENTS RECOMMENDING OR NOT 624 00:25:54,870 --> 00:25:56,004 RECOMMENDING WHATEVER. 625 00:25:56,004 --> 00:25:58,473 THAT INCLUDES -- THAT'S BESIDES 626 00:25:58,473 --> 00:26:01,543 THE STATEMENT THAT YOU WOULD 627 00:26:01,543 --> 00:26:03,378 MAKE ABOUT OUTSTANDING 628 00:26:03,378 --> 00:26:04,980 EXCELLENCE OR WHATEVER. 629 00:26:04,980 --> 00:26:07,115 >> YES, SO FOR JEREMY IS HE ON 630 00:26:07,115 --> 00:26:10,585 TRACK, FOR LAUREN IS SHE READY 631 00:26:10,585 --> 00:26:12,621 FOR CONSIDERATION. 632 00:26:12,621 --> 00:26:14,856 >> AND ALSO OBVIOUSLY THE BETTER 633 00:26:14,856 --> 00:26:18,860 SCORE, THE EASIER IT MAKES OUR 634 00:26:18,860 --> 00:26:19,061 LIFE. 635 00:26:19,061 --> 00:26:21,830 THEIR LIFE, ACTUALLY, NOT OURS. 636 00:26:21,830 --> 00:26:22,064 THANKS. 637 00:26:22,064 --> 00:26:28,236 GOOD QUESTION, SO THANK YOU. 638 00:26:28,236 --> 00:26:30,172 BY THE WAY, I WANT TO SAY THAT 639 00:26:30,172 --> 00:26:32,374 EVEN WITH ALL THE TROUBLES WE'VE 640 00:26:32,374 --> 00:26:35,444 HAD I APPRECIATE ALL OF YOUR 641 00:26:35,444 --> 00:26:42,284 ATTENDANCE TODAY VERY, VERY 642 00:26:42,284 --> 00:26:42,551 MUCH. 643 00:26:42,551 --> 00:26:45,253 AND SO WE'LL BE ABLE TO STAY 644 00:26:45,253 --> 00:26:48,323 THIS WAY, FOR THE WHOLE DAY NOW. 645 00:26:48,323 --> 00:26:53,362 IF YOU COME AND GO THROUGH THE 646 00:26:53,362 --> 00:26:56,898 DAY, THAT'S FINE, BUT NOW -- 647 00:26:56,898 --> 00:26:59,935 THANK YOU. 648 00:26:59,935 --> 00:27:00,969 SO IT'S 9:29. 649 00:27:00,969 --> 00:27:05,640 I DON'T SEE WHY WE DON'T START 650 00:27:05,640 --> 00:27:11,380 WITH PETER, CLEM IS REMOTE, SO 651 00:27:11,380 --> 00:27:14,783 HE CAN TURN ON HIS MIC AND SHARE 652 00:27:14,783 --> 00:27:15,617 HIS SLIDES. 653 00:27:15,617 --> 00:27:17,018 WHO IS SHARING THE SLIDES? 654 00:27:17,018 --> 00:27:21,757 AM I SHARING THE SLIDES? 655 00:27:21,757 --> 00:27:24,493 OR, SEAN, CAN YOU SHARE HIS 656 00:27:24,493 --> 00:27:24,726 SLIDES? 657 00:27:24,726 --> 00:27:25,794 I SEE YOU'RE ONLINE. 658 00:27:25,794 --> 00:27:30,399 THAT COULD PROBABLY BE EASIER. 659 00:27:30,399 --> 00:27:35,871 SEAN? 660 00:27:35,871 --> 00:27:37,706 >> GIVE ME ONE SECOND. 661 00:27:37,706 --> 00:27:41,042 YEAH, GIVE ME ONE SECOND. 662 00:27:41,042 --> 00:27:51,253 >> OKAY. 663 00:28:06,668 --> 00:28:10,472 >> DAVID, WHILE WE WAIT, I'M 664 00:28:10,472 --> 00:28:11,907 CURIOUS ABOUT OUR BOARD. 665 00:28:11,907 --> 00:28:14,876 HOW MANY MEMBERS ARE SUPPOSED TO 666 00:28:14,876 --> 00:28:18,280 HAVE, ESPECIALLY SINCE THE NEXT 667 00:28:18,280 --> 00:28:20,749 MEETING WE WILL HAVE FOUR PEOPLE 668 00:28:20,749 --> 00:28:25,187 TO REVIEW AND JUST LOSING THREE, 669 00:28:25,187 --> 00:28:28,657 THE PEOPLE THAT IS ACTUALLY 670 00:28:28,657 --> 00:28:28,890 LEAVING. 671 00:28:28,890 --> 00:28:32,093 >> SO, WE ARE GOING TO BE 672 00:28:32,093 --> 00:28:35,497 NOMINATING SIX NEW MEMBERS. 673 00:28:35,497 --> 00:28:38,500 THIS IS JUNE, THERE WILL BE SIX 674 00:28:38,500 --> 00:28:38,733 LEFT. 675 00:28:38,733 --> 00:28:42,771 WE'RE NOMINATING SIX MORE. 676 00:28:42,771 --> 00:28:44,773 WE CURRENTLY HAVE FOUR AND A 677 00:28:44,773 --> 00:28:47,409 HALF THAT HAVE ACCEPTED. 678 00:28:47,409 --> 00:28:48,977 I NEED ANOTHER PHONE CALL WITH 679 00:28:48,977 --> 00:28:49,411 ONE PERSON. 680 00:28:49,411 --> 00:28:51,546 SO THAT WILL BE FIVE AND ONE 681 00:28:51,546 --> 00:28:54,483 MORE PERSON WHO TURNED US DOWN 682 00:28:54,483 --> 00:28:59,287 BUT THAT'S MY -- WE'LL GET TO 683 00:28:59,287 --> 00:29:01,389 THE SIXTH PERSON. 684 00:29:01,389 --> 00:29:05,360 ONE PERSON AT A TIME, THEY ALL 685 00:29:05,360 --> 00:29:06,962 GET SUBMITTED AS A GROUP, THE 686 00:29:06,962 --> 00:29:08,630 CURRENT CHALLENGE RIGHT NOW IS 687 00:29:08,630 --> 00:29:13,168 WE DON'T KNOW HOW THIS HAPPENS, 688 00:29:13,168 --> 00:29:16,738 BECAUSE THE OFFICE WHO HANDLES 689 00:29:16,738 --> 00:29:18,807 THESE COMMITTEE MEETINGS AND 690 00:29:18,807 --> 00:29:22,244 THESE COMMITTEES IS NOT IN NLM 691 00:29:22,244 --> 00:29:22,477 ANYMORE. 692 00:29:22,477 --> 00:29:27,282 IT'S ABOVE NLM, SO THE PACKAGE 693 00:29:27,282 --> 00:29:27,916 WILL LEAVE NLM. 694 00:29:27,916 --> 00:29:30,719 I DON'T KNOW WHAT THE FUTURE -- 695 00:29:30,719 --> 00:29:32,687 HOW THAT'S GOING TO BE AND HOW 696 00:29:32,687 --> 00:29:33,688 QUICK THAT WILL BE. 697 00:29:33,688 --> 00:29:35,991 IF I SEE THAT IT'S NOT GOING TO 698 00:29:35,991 --> 00:29:39,728 BE QUICK, I'M GOING TO ADD IN AT 699 00:29:39,728 --> 00:29:44,266 LEAST THREE OR FOUR AD HOCS, 700 00:29:44,266 --> 00:29:47,369 WHICH IS MUCH EASIER TO DO, FOR 701 00:29:47,369 --> 00:29:47,602 SUPPORT. 702 00:29:47,602 --> 00:29:51,339 SO LET'S KEEP TALKING ABOUT IT. 703 00:29:51,339 --> 00:29:53,375 >> ALL RIGHT, CLEM, ARE YOU 704 00:29:53,375 --> 00:29:55,110 READY TO GO? 705 00:29:55,110 --> 00:29:59,381 IF YOU CAN SAY SOMETHING TO 706 00:29:59,381 --> 00:30:05,420 CHECK YOUR MIC. 707 00:30:05,420 --> 00:30:06,721 CLEM? 708 00:30:06,721 --> 00:30:11,693 >> LOOKS LIKE ON MY END HE MUTED 709 00:30:11,693 --> 00:30:12,761 HIS MIC AGAIN. 710 00:30:12,761 --> 00:30:15,497 DR. MCDONALD, IF YOU CAN UNMUTE 711 00:30:15,497 --> 00:30:16,398 YOUR MIC. 712 00:30:16,398 --> 00:30:18,867 AND THAT WOULD BE AT THE VERY 713 00:30:18,867 --> 00:30:22,203 TOP TOOL BAR. 714 00:30:22,203 --> 00:30:24,172 THERE'S A RED BUTTON THAT SAYS 715 00:30:24,172 --> 00:30:26,708 LEAVE, TO THE LEFT. 716 00:30:26,708 --> 00:30:32,113 THERE YOU GO. 717 00:30:32,113 --> 00:30:34,249 >> THERE YOU GO. 718 00:30:34,249 --> 00:30:35,650 PRESS IT ONCE. 719 00:30:35,650 --> 00:30:37,285 THANK YOU. 720 00:30:37,285 --> 00:30:39,921 CLEM? 721 00:30:39,921 --> 00:30:42,591 WE CAN HEAR YOU. 722 00:30:42,591 --> 00:30:43,258 >> YOU CAN HEAR ME? 723 00:30:43,258 --> 00:30:43,625 >> YES. 724 00:30:43,625 --> 00:30:50,832 >> OH, GOOD. 725 00:30:50,832 --> 00:30:53,335 >> PLEASE BEGIN. 726 00:30:53,335 --> 00:30:53,635 >> PARDON? 727 00:30:53,635 --> 00:30:59,107 >> YOU'RE ON. 728 00:30:59,107 --> 00:31:01,743 >> I'M NOT FOLLOWING. 729 00:31:01,743 --> 00:31:02,844 >> IT'S TIME FOR YOUR 730 00:31:02,844 --> 00:31:03,378 PRESENTATION. 731 00:31:03,378 --> 00:31:04,746 SOMEBODY OTHER THAN YOU, I 732 00:31:04,746 --> 00:31:08,450 BELIEVE, IS SHARING THE SLIDES. 733 00:31:08,450 --> 00:31:10,352 SO YOU'LL NEED TO LET THEM KNOW 734 00:31:10,352 --> 00:31:12,654 WHEN TO ADVANCE YOUR SLIDES BUT 735 00:31:12,654 --> 00:31:15,590 GO AHEAD AND START YOUR 736 00:31:15,590 --> 00:31:16,191 PRESENTATION. 737 00:31:16,191 --> 00:31:18,960 >> OKAY, SEAN, ARE YOU DOING THE 738 00:31:18,960 --> 00:31:19,194 SLIDES? 739 00:31:19,194 --> 00:31:21,029 >> YES, HE'S GOING TO -- 740 00:31:21,029 --> 00:31:22,597 >> YES, I'M SHARING THE SLIDES. 741 00:31:22,597 --> 00:31:23,064 >> OKAY. 742 00:31:23,064 --> 00:31:25,767 ALL RIGHT. 743 00:31:25,767 --> 00:31:33,675 NEXT SLIDE. 744 00:31:33,675 --> 00:31:35,276 SO THIS IS OUR AGENDA. 745 00:31:35,276 --> 00:31:38,680 WE'LL FIRST TALK ABOUT THE 746 00:31:38,680 --> 00:31:41,616 RESEARCH TEAM, BACKGROUND AND 747 00:31:41,616 --> 00:31:43,918 OBJECTIVES, WORKFLOW, RESOURCES 748 00:31:43,918 --> 00:31:45,887 AND INFRASTRUCTURE, HIGHLIGHTS, 749 00:31:45,887 --> 00:31:52,260 FUTURE DIRECTIONS, SUMMARY OF 750 00:31:52,260 --> 00:31:52,560 DISCUSSION. 751 00:31:52,560 --> 00:31:55,230 NEXT SLIDE. 752 00:31:55,230 --> 00:32:05,640 THE TEAM IS ME, A SCIENTIST, -- 753 00:32:05,640 --> 00:32:08,543 >> HE MUTED AGAIN. 754 00:32:08,543 --> 00:32:09,477 >> I'M SORRY? 755 00:32:09,477 --> 00:32:10,211 >> GO AHEAD. 756 00:32:10,211 --> 00:32:14,049 >> CAN YOU HEAR ME? 757 00:32:14,049 --> 00:32:15,583 >> YES. 758 00:32:15,583 --> 00:32:16,851 >> OKAY. 759 00:32:16,851 --> 00:32:23,491 THE LISTER HILL DIRECTOR. 760 00:32:23,491 --> 00:32:25,326 NEXT SLIDE. 761 00:32:25,326 --> 00:32:29,064 NOW, WE HAD A MARVELOUS STUDENT 762 00:32:29,064 --> 00:32:31,232 WITH US DURING THIS LAST TOUR, 763 00:32:31,232 --> 00:32:34,969 WHO IS NOW A RESIDENT PHYSICIAN 764 00:32:34,969 --> 00:32:36,771 IN BRIGHAM AND WOMEN'S HOSPITAL, 765 00:32:36,771 --> 00:32:42,744 HE DID A STUDY OF EFFECTIVENESS 766 00:32:42,744 --> 00:32:53,288 OF DIRECT ORAL ANTI-COAGULATION 767 00:32:58,760 --> 00:33:01,930 WITH RIVAROXABAN VERSUS APAXABAN 768 00:33:01,930 --> 00:33:06,634 AND RISK OF HOSPITALIZATION FOR 769 00:33:06,634 --> 00:33:06,935 BLEEDING. 770 00:33:06,935 --> 00:33:08,737 THE BACKGROUND, OLDER AND 771 00:33:08,737 --> 00:33:14,275 CHRONICALLY ILL PEOPLE ARE OFTEN 772 00:33:14,275 --> 00:33:15,777 EXCLUDED FROM RANDOMIZED TRIAL, 773 00:33:15,777 --> 00:33:18,546 POST MARKING SURVEILLANCE HELPS 774 00:33:18,546 --> 00:33:23,585 UNCOVER REAL WORLD DRUG EFFECT, 775 00:33:23,585 --> 00:33:25,053 AND WE NEED RETROSPECTIVE 776 00:33:25,053 --> 00:33:25,453 REVIEW. 777 00:33:25,453 --> 00:33:27,655 THAT'S WHAT OUR GAME HAS BEEN, 778 00:33:27,655 --> 00:33:29,491 USING MOSTLY MEDICARE AS THE 779 00:33:29,491 --> 00:33:33,995 LARGE DATASET. 780 00:33:33,995 --> 00:33:34,963 NEXT SLIDE. 781 00:33:34,963 --> 00:33:38,600 AND WE USE IT TO EXTEND DRUG 782 00:33:38,600 --> 00:33:42,403 SAFETY FINDING AND TO 783 00:33:42,403 --> 00:33:43,772 INVESTIGATE EMERGING AND 784 00:33:43,772 --> 00:33:46,307 CONTROVERSIAL DRUG EFFECTS, AND 785 00:33:46,307 --> 00:33:48,309 WE DISSEMINATED POLICY AND 786 00:33:48,309 --> 00:33:53,081 CLINICAL GUIDANCE ON MEDICATION 787 00:33:53,081 --> 00:33:53,448 USE. 788 00:33:53,448 --> 00:33:55,817 NEXT SLIDE. 789 00:33:55,817 --> 00:34:00,054 NEXT SLIDE. 790 00:34:00,054 --> 00:34:02,323 SO, HYPOTHESIS, TEST IF TARGET 791 00:34:02,323 --> 00:34:05,460 MEDICATIONS ARE SAFE, IF THEY 792 00:34:05,460 --> 00:34:07,495 HAVE ANY BENEFITS. 793 00:34:07,495 --> 00:34:09,564 AND PER DATA WE USED 794 00:34:09,564 --> 00:34:11,599 PRESCRIPTION AND MEDICAL RECORDS 795 00:34:11,599 --> 00:34:15,336 AND METHOD WE USED SURVIVAL 796 00:34:15,336 --> 00:34:21,442 ANALYSIS AND WE TEST HYPOTHESIS 797 00:34:21,442 --> 00:34:23,111 USING COX REGRESSION WITH TIME 798 00:34:23,111 --> 00:34:25,847 VARYING COVARIATES AND TIME 799 00:34:25,847 --> 00:34:34,622 VARYING PROPENSITY SCORE. 800 00:34:34,622 --> 00:34:37,559 NEXT SLIDE. 801 00:34:37,559 --> 00:34:41,162 OUR DATA SOURCE IS MEDICARE VRDC 802 00:34:41,162 --> 00:34:44,065 PROVIDING ACCESS TO AVAILABLE 803 00:34:44,065 --> 00:34:50,438 RECORDS OF 100% OF MEDICARE 804 00:34:50,438 --> 00:34:54,776 ENROLLEES, INCLUDES ENROLLMENT, 805 00:34:54,776 --> 00:34:55,977 DEMOGRAPHICS, CONDITIONS, 806 00:34:55,977 --> 00:34:57,879 SPENDING DATA 1999 TO 2022. 807 00:34:57,879 --> 00:35:03,585 PART A HOSPITAL INSURANCE CLAIMS 808 00:35:03,585 --> 00:35:06,955 INCLUDES 1999 TO 2022. 809 00:35:06,955 --> 00:35:08,656 PART B MEDICAL CLAIMS ARE 810 00:35:08,656 --> 00:35:12,727 PROVIDED BY THE SAME TIME 811 00:35:12,727 --> 00:35:15,930 WINDOW, AND PART C MEDICARE 812 00:35:15,930 --> 00:35:18,366 ADVANTAGE CLAIMS DATA INCLUDED 813 00:35:18,366 --> 00:35:21,336 IN 2015 TO 2022. 814 00:35:21,336 --> 00:35:26,441 AND PART D PRESCRIPTION DRUG 815 00:35:26,441 --> 00:35:36,985 INSURANCE, CLAIMS DATA, 2006 TO 816 00:35:40,788 --> 00:35:41,289 2022. 817 00:35:41,289 --> 00:35:44,959 NEXT SLIDE. 818 00:35:44,959 --> 00:35:49,130 A VERY LARGE DATASET FROM CMS 819 00:35:49,130 --> 00:35:51,466 CAN SUPPORT EPIDEMIOLOGIC AND 820 00:35:51,466 --> 00:35:56,137 SAFETY STUDIES, CRITICAL FOR 821 00:35:56,137 --> 00:35:58,106 REAL WORLD EVIDENCE GENERATION. 822 00:35:58,106 --> 00:35:59,941 EXCUSE ME. 823 00:35:59,941 --> 00:36:07,849 NEXT SLIDE. 824 00:36:07,849 --> 00:36:11,052 NOW, OUR STUDY, A NUMBER OF 825 00:36:11,052 --> 00:36:13,354 THEM, HORMONE REPLACEMENT 826 00:36:13,354 --> 00:36:16,057 THERAPY, USE OF TESTOSTERONE, 827 00:36:16,057 --> 00:36:26,467 ESTROGEN AND PROGESTIN, 828 00:36:27,535 --> 00:36:28,002 MEDICATION HYPERTENSION, 829 00:36:28,002 --> 00:36:29,771 DIABETES, ANTI-COAGULATION, 830 00:36:29,771 --> 00:36:31,940 ANTIBIOTICS AND OTHERS AND 831 00:36:31,940 --> 00:36:33,841 COVID-19 TREATMENT, WE ALSO DID 832 00:36:33,841 --> 00:36:37,045 SOME AREAS SOME STUDY OF LOINC 833 00:36:37,045 --> 00:36:40,782 AS A STANDARD CODE FOR THAT 834 00:36:40,782 --> 00:36:47,522 CLINICAL TEST. 835 00:36:47,522 --> 00:36:53,127 WE PUBLISHED IN MENOPAUSE JAMA, 836 00:36:53,127 --> 00:36:57,165 PLOS MEDICINE, PLOS ONE, AND 837 00:36:57,165 --> 00:37:02,770 OTHERS, BRITISH MEDICAL JOURNAL, 838 00:37:02,770 --> 00:37:04,372 THROMBOSIS, REAL WORLD OUTCOMES, 839 00:37:04,372 --> 00:37:08,743 HEALTH DISPARITIES, ET CETERA. 840 00:37:08,743 --> 00:37:09,043 NEXT SLIDE. 841 00:37:09,043 --> 00:37:18,686 THE JOURNALS WE PUBLISHED IN. 842 00:37:18,686 --> 00:37:22,156 NOW, BACKGROUND, MENOPAUSE, END 843 00:37:22,156 --> 00:37:25,426 OF REPRODUCTION YEARS, HORMONES, 844 00:37:25,426 --> 00:37:27,061 SYMPTOMS INCLUDE HOT FLASHES, 845 00:37:27,061 --> 00:37:29,564 WEIGHT GAIN, ET CETERA, AND 846 00:37:29,564 --> 00:37:32,066 LONG-TERM EFFECTS IF UNTREATED, 847 00:37:32,066 --> 00:37:33,201 PROBLEMS WITH BONE HEALTH, 848 00:37:33,201 --> 00:37:37,171 CARDIOVASCULAR HEALTH, ET 849 00:37:37,171 --> 00:37:37,805 CETERA. 850 00:37:37,805 --> 00:37:41,209 THE WHI HORMONE TRIALS, IT 851 00:37:41,209 --> 00:37:43,511 SHOWED ESTROGEN AND PROGESTIN 852 00:37:43,511 --> 00:37:46,014 THERAPY INCREASED BREAST CANCER, 853 00:37:46,014 --> 00:37:49,283 STROKE, CARDIAC, CORONARY HEART 854 00:37:49,283 --> 00:37:51,519 DISEASE, ESTROGEN ALONE THERAPY 855 00:37:51,519 --> 00:37:58,526 INCREASED STROKE AND DEMENTIA, 856 00:37:58,526 --> 00:37:59,360 AND WAS STOPPED PREMATURELY 857 00:37:59,360 --> 00:38:05,566 BETWEEN 2002 AND 2004. 858 00:38:05,566 --> 00:38:07,368 NEXT SLIDE. 859 00:38:07,368 --> 00:38:17,645 SO WHI STUDY STUDIED ONE 860 00:38:17,645 --> 00:38:20,715 PREPARATION, ORAL .625 861 00:38:20,715 --> 00:38:24,285 MILLIGRAMS, AND ORAL -- THE SAME 862 00:38:24,285 --> 00:38:31,959 DRUG PLUS, AND THE TYPE OF 863 00:38:31,959 --> 00:38:34,128 ESTROGEN WAS CONJUGATED, EQUINE 864 00:38:34,128 --> 00:38:40,468 ESTROGEN, TYPE OF PROGESTIN WAS 865 00:38:40,468 --> 00:38:45,840 MEDROXY, ORAL ROUTE, STRENGTH 866 00:38:45,840 --> 00:38:49,977 .625, AND A BLACK BOX WARNING ON 867 00:38:49,977 --> 00:38:57,518 ARE FROM THE FDA ON ALL 868 00:38:57,518 --> 00:38:58,319 ESTROGEN, QUESTION EFFECTS 869 00:38:58,319 --> 00:38:59,320 ACROSS ROUTES OF ADMINISTRATION 870 00:38:59,320 --> 00:39:01,389 AND STRENGTH. 871 00:39:01,389 --> 00:39:08,029 NEXT SLIDE. 872 00:39:08,029 --> 00:39:09,163 EXCUSE ME. 873 00:39:09,163 --> 00:39:12,467 WE USED PRESCRIPTION AND MEDICAL 874 00:39:12,467 --> 00:39:17,772 AND COUNTER RECORDS FROM 875 00:39:17,772 --> 00:39:18,806 TEN-PLUS MILLION POSTMENOPAUSAL 876 00:39:18,806 --> 00:39:23,144 WOMAN AND HORMONE THERAPY 877 00:39:23,144 --> 00:39:24,846 CLASSIFIED AS ESTRADIOL, AND 878 00:39:24,846 --> 00:39:30,852 OTHERS, TYPE OF PROGESTIN, 879 00:39:30,852 --> 00:39:38,926 ADMINISTRATION ROUTE ORAL, 880 00:39:38,926 --> 00:39:41,696 VAGINAL, TRANSDERMAL AND VILE. 881 00:39:41,696 --> 00:39:43,764 IN TOTAL LOOKED AT 40 DISTINCT 882 00:39:43,764 --> 00:39:54,275 HORMONE THERAPY PREPARATIONS. 883 00:39:58,079 --> 00:40:00,581 NEXT SLIDE. 884 00:40:00,581 --> 00:40:03,084 INCLUDING ALL-CAUSE MORTALITY, 885 00:40:03,084 --> 00:40:06,554 DEMENTIA, CANCERS, BREAST, LUNG, 886 00:40:06,554 --> 00:40:08,723 COLORECTAL, OVARIAN. 887 00:40:08,723 --> 00:40:11,893 AND SIX CARDIOVASCULAR EVENTS, 888 00:40:11,893 --> 00:40:16,898 ISCHEMIC HEART DISEASE, CORONARY 889 00:40:16,898 --> 00:40:20,535 HEART -- CONGESTIVE HEART 890 00:40:20,535 --> 00:40:22,937 FAILURE, VTE, STROKE, AMI. 891 00:40:22,937 --> 00:40:24,539 NEXT SLIDE. 892 00:40:24,539 --> 00:40:28,743 WE USED A TIME TO EVENT DATA 893 00:40:28,743 --> 00:40:30,478 ANALYSIS INCLUDING COX 894 00:40:30,478 --> 00:40:32,713 REGRESSION OF TEN-PLUS MILLION 895 00:40:32,713 --> 00:40:38,986 FOR 13 PATIENT OUTCOMES, 896 00:40:38,986 --> 00:40:39,921 ADJUSTING TIME DEPENDENT, 897 00:40:39,921 --> 00:40:44,725 INCLUDING 40 HORMONE THERAPY 898 00:40:44,725 --> 00:40:49,363 PREPARATIONS, DEMOGRAPHICS, TIME 899 00:40:49,363 --> 00:40:51,199 FIXED, SOCIOECONOMICS, INCOME, 900 00:40:51,199 --> 00:40:54,001 REGIONS, NORTH, SOUTH, ET 901 00:40:54,001 --> 00:40:56,871 CETERA, RURAL RESIDENCY, 49 902 00:40:56,871 --> 00:41:00,775 CHRONIC CONDITIONS, USED 903 00:41:00,775 --> 00:41:08,783 PROPENSITY SCORE ADJUSTMENT. 904 00:41:08,783 --> 00:41:09,283 NEXT SLIDE. 905 00:41:09,283 --> 00:41:12,587 RESULTS -- EFFECTS VARIED BY 906 00:41:12,587 --> 00:41:17,925 TYPES, ROUTES, STRENGTH OF THE 907 00:41:17,925 --> 00:41:19,594 HORMONE THERAPY. 908 00:41:19,594 --> 00:41:22,230 ESTROGEN THERAPY ASSOCIATED WITH 909 00:41:22,230 --> 00:41:25,733 SIGNIFICANT RISK OF REDUCTION, 910 00:41:25,733 --> 00:41:28,736 THAT'S IMPORTANT, REDUCTION, IN 911 00:41:28,736 --> 00:41:32,740 ALL-CAUSE MORTALITY, DECLINE OF 912 00:41:32,740 --> 00:41:37,612 19%, HAZARD RATIO 0.81. 913 00:41:37,612 --> 00:41:39,714 BREAST CANCER REDUCTION 16%, 914 00:41:39,714 --> 00:41:42,516 LUNG 13% REDUCTION IN LUNG 915 00:41:42,516 --> 00:41:45,386 CANCER, 12% REDUCTION IN 916 00:41:45,386 --> 00:41:47,922 COLORECTAL CANCER, 5% REDUCTION 917 00:41:47,922 --> 00:41:50,625 IN HEART FAILURE, 3% REDUCTION 918 00:41:50,625 --> 00:42:00,768 IN VTE, 4% IN A-FIB, 11% AMI, 2% 919 00:42:00,768 --> 00:42:01,402 DEMENTIA. 920 00:42:01,402 --> 00:42:04,772 EPT WAS LINKED TO SIGNIFICANT 921 00:42:04,772 --> 00:42:12,747 INCREASED RISK IN BREAST CANCER, 922 00:42:12,747 --> 00:42:15,383 E+ PROGESTIN INCREASED 10%, 923 00:42:15,383 --> 00:42:19,620 INCLEESED E+ PROGESTERONE 924 00:42:19,620 --> 00:42:22,590 INCREASE OF 19%. 925 00:42:22,590 --> 00:42:26,294 BUT MITIGATED BY LOW VAGINAL -- 926 00:42:26,294 --> 00:42:29,930 CAN'T READ THAT ONE, 927 00:42:29,930 --> 00:42:36,370 TRANSDERMAL, AND PROGESTIN, E2+ 928 00:42:36,370 --> 00:42:38,773 PROGESTIN. 929 00:42:38,773 --> 00:42:39,407 NEXT SLIDE. 930 00:42:39,407 --> 00:42:45,479 SO, THIS PAPER ABOUT THE USE OF 931 00:42:45,479 --> 00:42:48,949 MENOPAUSAL HORMONE THERAPY AGE 932 00:42:48,949 --> 00:42:52,353 5, EFFECT OF OUTCOMES BY TYPES 933 00:42:52,353 --> 00:42:55,556 OF ROUTES HAD A GOOD SPLASH, TWO 934 00:42:55,556 --> 00:42:58,059 MONTHS AFTER PUBLICATION RANKED 935 00:42:58,059 --> 00:43:02,530 IN THE TOP 5% OF ALL 26 MILLION 936 00:43:02,530 --> 00:43:04,765 RESEARCH OUTPUTS, COVERED 22 937 00:43:04,765 --> 00:43:10,438 DIFFERENT NEWS OUTLETS, THAT WAS 938 00:43:10,438 --> 00:43:10,604 FUN. 939 00:43:10,604 --> 00:43:14,809 NEXT SLIDE. 940 00:43:14,809 --> 00:43:19,847 COVID-19 STUDY, SO THIS IS A 941 00:43:19,847 --> 00:43:25,686 SMALL -- WE DEVELOPED A VACCINE 942 00:43:25,686 --> 00:43:29,790 PRIORITIZATION MODEL, PUBLISHED 943 00:43:29,790 --> 00:43:34,929 IN MEDR 2021, AVAILABILITY OF 944 00:43:34,929 --> 00:43:37,598 THE COVID-19 VACCINE, LIMITED 945 00:43:37,598 --> 00:43:39,266 AVAILABILITY, SO WE WANTED TO 946 00:43:39,266 --> 00:43:40,301 FIGURE OUT WHICH POPULATION 947 00:43:40,301 --> 00:43:42,503 GROUP WOULD GAIN THE MOST 948 00:43:42,503 --> 00:43:43,871 BENEFIT FROM VACCINATION. 949 00:43:43,871 --> 00:43:47,575 WE USED MONTHLY DATA FROM THE 950 00:43:47,575 --> 00:43:50,044 NATIONAL CENTER FOR HEALTH 951 00:43:50,044 --> 00:43:56,250 STATISTICS ABOUT COVID-19 952 00:43:56,250 --> 00:44:01,756 DEATHS, USING CD10 CODE UO 7.1. 953 00:44:01,756 --> 00:44:02,056 NEXT SLIDE. 954 00:44:02,056 --> 00:44:06,026 WE USED -- THE METHOD WAS 955 00:44:06,026 --> 00:44:08,763 LOGISTIC REGRESSION ANALYSIS TO 956 00:44:08,763 --> 00:44:11,699 ASSESS VULNERABILITY TO COVID 957 00:44:11,699 --> 00:44:13,234 MORTALITY, COVID-19 MORTALITY, 958 00:44:13,234 --> 00:44:17,571 BY AGE, RACE, SEX. 959 00:44:17,571 --> 00:44:18,739 RESULTS, VACCINATING THE TOP 17 960 00:44:18,739 --> 00:44:23,677 HIGH RISK GROUPS OF POPULATION 961 00:44:23,677 --> 00:44:25,579 WHICH INCLUDED MALES AGED 85, 962 00:44:25,579 --> 00:44:27,515 BLACK AND HISPANIC, COULD 963 00:44:27,515 --> 00:44:33,521 PREVENT 47% OF COVID-19 DEATHS. 964 00:44:33,521 --> 00:44:35,990 SO, THIS INFORMED EARLY VACCINE 965 00:44:35,990 --> 00:44:43,597 DISTRIBUTION STRATEGIES. 966 00:44:43,597 --> 00:44:44,632 NEXT SLIDE. 967 00:44:44,632 --> 00:44:46,100 EXCUSE ME. 968 00:44:46,100 --> 00:44:47,768 SOME BACKGROUND, INCIDENCE OF 969 00:44:47,768 --> 00:44:49,503 LONG COVID IN THE ELDERLY IS 970 00:44:49,503 --> 00:44:51,405 DIFFICULT TO ESTIMATE. 971 00:44:51,405 --> 00:44:54,341 SYMPTOMS MAY BE CONFUSED WITH 972 00:44:54,341 --> 00:44:56,944 OTHER CHRONIC CONDITIONS. 973 00:44:56,944 --> 00:45:01,582 SO PROCESS WAS INCIDENCE 974 00:45:01,582 --> 00:45:04,752 BASED -- SPECIFIC DIAGNOSIS ON 975 00:45:04,752 --> 00:45:06,187 COVID MAYBE UNDERREPORTED, AND 976 00:45:06,187 --> 00:45:07,822 POST-COVID HEALTH CARE 977 00:45:07,822 --> 00:45:11,358 UTILIZATION PATTERNS ARE LIKELY 978 00:45:11,358 --> 00:45:18,432 TO DIFFER FROM THOSE OF 979 00:45:18,432 --> 00:45:18,699 INFLUENZA. 980 00:45:18,699 --> 00:45:19,633 NEXT SLIDE. 981 00:45:19,633 --> 00:45:23,737 WE USED AS OUR COVID-19 COHORT 982 00:45:23,737 --> 00:45:28,742 WE USED CMS MEDICARE ENCOUNTER 983 00:45:28,742 --> 00:45:31,645 RECORDS OF APRIL 2020, JUNE 984 00:45:31,645 --> 00:45:34,682 2021, AND INFLUENZA COHORT USED 985 00:45:34,682 --> 00:45:37,952 CMS MEDICAL ENCOUNTER RECORDS 986 00:45:37,952 --> 00:45:41,755 FROM OCTOBER-MAY IN 2018-2019. 987 00:45:41,755 --> 00:45:48,762 SO, WE DEFINED LONG COVID BASED 988 00:45:48,762 --> 00:45:52,766 ON ICD-10 EM B94.8 SEQUELAE, 989 00:45:52,766 --> 00:45:56,070 OTHER SPECIFIC INFECTIONS AND 990 00:45:56,070 --> 00:45:56,637 PARASITIC DISEASES, SYMPTOM 991 00:45:56,637 --> 00:46:00,741 BASED, W.H.O. CLINICAL 992 00:46:00,741 --> 00:46:01,709 DEFINITIONS, DESCRIBED 11 993 00:46:01,709 --> 00:46:03,577 SYMPTOMS THAT OCCURRED WITHIN 3 994 00:46:03,577 --> 00:46:04,912 MONTHS OF INFECTION, AND LAST 995 00:46:04,912 --> 00:46:07,081 FOR GREATER THAN 2 MONTHS. 996 00:46:07,081 --> 00:46:12,586 THAT WAS OUR DEFINITION OF LONG 997 00:46:12,586 --> 00:46:12,786 COVID. 998 00:46:12,786 --> 00:46:13,087 NEXT SLIDE. 999 00:46:13,087 --> 00:46:17,725 NEXT SLIDE. 1000 00:46:17,725 --> 00:46:20,761 SO RESULTS IMPLICATION, CODE 1001 00:46:20,761 --> 00:46:27,134 BASED VERSUS SYMPTOM BASED LONG 1002 00:46:27,134 --> 00:46:31,205 COVID WAS REALLY OVERBALANCED 1003 00:46:31,205 --> 00:46:34,074 WITH THE SYMPTOM BASE. 1004 00:46:34,074 --> 00:46:38,112 2.6 VERSUS 30%, 2.6% VERSUS 30% 1005 00:46:38,112 --> 00:46:40,114 AMONG HOSPITALIZED COVID 1006 00:46:40,114 --> 00:46:41,081 PATIENTS UNDERREPORTED. 1007 00:46:41,081 --> 00:46:43,851 THE HIGHER BURDEN FROM LONG 1008 00:46:43,851 --> 00:46:46,320 COVID COMPARED TO POST INFLUENZA 1009 00:46:46,320 --> 00:46:49,924 CONDITIONS, SO THE INCIDENCE OF 1010 00:46:49,924 --> 00:46:51,125 DYSPNEA, FATIGUE, PALPITATIONS, 1011 00:46:51,125 --> 00:46:55,829 LOSS OF TASTE, SMELL, AND 1012 00:46:55,829 --> 00:46:56,330 NEUROCOGNITIVE SYMPTOMS, 1013 00:46:56,330 --> 00:46:59,233 INCIDENCE OF HOSPITALIZATION AND 1014 00:46:59,233 --> 00:47:01,802 OUTPATIENT VISITS INCREASED 1015 00:47:01,802 --> 00:47:02,436 ALSO. 1016 00:47:02,436 --> 00:47:04,071 HIGHLIGHTS, THESE STUDIES 1017 00:47:04,071 --> 00:47:07,241 HIGHLIGHT THE NEED FOR VIGILANT 1018 00:47:07,241 --> 00:47:08,742 MONITORING AND TAILORED 1019 00:47:08,742 --> 00:47:12,580 HEALTHCARE STRATEGIES FOR 1020 00:47:12,580 --> 00:47:13,847 ELDERLY INDIVIDUALS RECOVERING 1021 00:47:13,847 --> 00:47:19,320 FROM COVID-19. 1022 00:47:19,320 --> 00:47:20,955 NEXT SLIDE. 1023 00:47:20,955 --> 00:47:24,992 SO, SOME BACKGROUND, THERE'S NO 1024 00:47:24,992 --> 00:47:31,465 FDA-APPROVED MEDICATIONS FOR THE 1025 00:47:31,465 --> 00:47:34,468 PREVENTION OF LONG COVID. 1026 00:47:34,468 --> 00:47:38,539 THE WE HYPOTHESIZE OR ASKED 1027 00:47:38,539 --> 00:47:44,912 WHETHER TWO COVID-19 TREATMENTS, 1028 00:47:44,912 --> 00:47:47,982 PAXLOVID AND LAGEVRIO ARE 1029 00:47:47,982 --> 00:47:49,683 SOCIALED WITH REDUCED RISK, 1030 00:47:49,683 --> 00:47:54,421 LOOKED AT CMS PRESCRIPTION AND 1031 00:47:54,421 --> 00:47:55,756 MEDICAL ENCOUNTER RECORDS IN 1032 00:47:55,756 --> 00:47:59,827 2021 AND 2022. 1033 00:47:59,827 --> 00:48:01,095 NEXT SLIDE. 1034 00:48:01,095 --> 00:48:04,531 METHODS, WE ONLY CONSIDERED TWO 1035 00:48:04,531 --> 00:48:07,134 COVID TREATMENTS TAKEN WITHIN 1036 00:48:07,134 --> 00:48:09,103 FIVE -- PLUS OR MINUS FIVE DAYS 1037 00:48:09,103 --> 00:48:11,672 OF COVID-19 DIAGNOSIS WAS MOST 1038 00:48:11,672 --> 00:48:14,508 EFFECTIVE IN THAT WINDOW. 1039 00:48:14,508 --> 00:48:20,848 WE USED COX REGRESSION ANALYSIS 1040 00:48:20,848 --> 00:48:26,053 OF LONG COVID RISK ON -- SORRY. 1041 00:48:26,053 --> 00:48:31,492 EFFECTS ON LONG COVID RISK, 1042 00:48:31,492 --> 00:48:34,328 ADJUSTING FOR DEMOGRAPHICS, 1043 00:48:34,328 --> 00:48:34,862 SOCIOECONOMIC, GEOGRAPHIC 1044 00:48:34,862 --> 00:48:40,234 REGIONS, 51 CHRONIC CONDITIONS. 1045 00:48:40,234 --> 00:48:42,503 RESULTS, PAXLOVID DECREASED PCC 1046 00:48:42,503 --> 00:48:48,242 RISK BY 13%. 1047 00:48:48,242 --> 00:48:53,280 AND LAGEVRIO DECREASED BY 8%, SO 1048 00:48:53,280 --> 00:48:54,782 THIS ARGUES FOR USING ONE OR 1049 00:48:54,782 --> 00:48:57,785 BOTH OF THESE DRUGS TO MITIGATE 1050 00:48:57,785 --> 00:49:01,221 THE RISK GIVEN THE LACK OF 1051 00:49:01,221 --> 00:49:07,461 PCC-SPECIFIC TREATMENT. 1052 00:49:07,461 --> 00:49:08,762 NEXT SLIDE. 1053 00:49:08,762 --> 00:49:10,798 THIS IS A DIFFERENT SORT OF 1054 00:49:10,798 --> 00:49:11,298 POINT. 1055 00:49:11,298 --> 00:49:16,203 WE DID A LITTLE STUDY ON LOINC, 1056 00:49:16,203 --> 00:49:18,572 LOGICAL OBSERVATION IDENTIFIER, 1057 00:49:18,572 --> 00:49:21,008 CHANGING CODES, USED TO IDENTIFY 1058 00:49:21,008 --> 00:49:26,447 TESTS AND MEASUREMENTS IN 1059 00:49:26,447 --> 00:49:29,483 HEALTHCARE, THE AIM WAS ASSIST 1060 00:49:29,483 --> 00:49:32,519 ACCURACY OF LAB CODE MAPPINGS TO 1061 00:49:32,519 --> 00:49:35,489 LOINC, DEVELOPED AN ALGORITHM TO 1062 00:49:35,489 --> 00:49:36,724 AUTOMATICALLY CORRECT ANY 1063 00:49:36,724 --> 00:49:38,392 ERRORS, MANY -- MOST ERRORS. 1064 00:49:38,392 --> 00:49:42,763 NEXT SLIDE. 1065 00:49:42,763 --> 00:49:47,868 THAT WAS PUBLISHED IN 2023. 1066 00:49:47,868 --> 00:49:50,137 WE USED 183 MILLION TEST 1067 00:49:50,137 --> 00:49:57,644 RESULTS, 4,000 -- OVER 4,000 1068 00:49:57,644 --> 00:49:58,979 MEASUREMENTS FROM PCORnet, 1069 00:49:58,979 --> 00:50:01,815 OVER 60 DATA MARKS, USED 1070 00:50:01,815 --> 00:50:03,016 SOFTWARE TELLS AND MANUAL REVIEW 1071 00:50:03,016 --> 00:50:07,087 TO ESTIMATE RATE OF LOINC 1072 00:50:07,087 --> 00:50:09,189 MAPPING ERRORS. 1073 00:50:09,189 --> 00:50:12,960 MAPPING ERROR RATE WAS 4.6% IN 1074 00:50:12,960 --> 00:50:15,696 PCORnet, DECREASED TO 0.1% 1075 00:50:15,696 --> 00:50:17,498 AFTER THE APPLICATION OF 1076 00:50:17,498 --> 00:50:20,033 AUTOMATIC DETECTION AND 1077 00:50:20,033 --> 00:50:22,669 CORRECTION ALGORITHM. 1078 00:50:22,669 --> 00:50:26,406 AND THIS ALL SUPPORTS DATA 1079 00:50:26,406 --> 00:50:29,376 INTEROPERABILITY AND QUALITY 1080 00:50:29,376 --> 00:50:32,780 ATURNS. -- ASSURANCE, PUBLISHED 1081 00:50:32,780 --> 00:50:35,482 IN JANUARY 2023. 1082 00:50:35,482 --> 00:50:39,253 SO GOING FORWARD, WE PLAN TO 1083 00:50:39,253 --> 00:50:41,755 CONTINUE THIS LOINC SUBSAFETY 1084 00:50:41,755 --> 00:50:43,624 ACROSS POPULATIONS, WE'RE GOING 1085 00:50:43,624 --> 00:50:47,127 TO UPDATE THE MENOPAUSAL HORMONE 1086 00:50:47,127 --> 00:50:51,832 THERAPY STUDIES AND USING 13 1087 00:50:51,832 --> 00:50:52,866 MILLION POST-MENOPAUSAL WOMEN 1088 00:50:52,866 --> 00:50:56,203 DATA, WE WANT TO INFORM POLICY 1089 00:50:56,203 --> 00:50:57,337 AND CLINICAL GUIDANCE ON 1090 00:50:57,337 --> 00:51:05,512 MEDICATION USE IN THE ELDERLY. 1091 00:51:05,512 --> 00:51:08,749 THIS IS THE LAST ONE. 1092 00:51:08,749 --> 00:51:13,420 NEXT SLIDE. 1093 00:51:13,420 --> 00:51:14,588 WHATEVER IT IS. 1094 00:51:14,588 --> 00:51:20,761 FUTURE PLANS, WE HAVE SOME IDEAS 1095 00:51:20,761 --> 00:51:24,765 FOR METHODOLOGIC INNOVATION. 1096 00:51:24,765 --> 00:51:27,100 THE TEAM IS DEFINING A METHOD TO 1097 00:51:27,100 --> 00:51:29,937 INTEGRATE DEEP LEARNING WITH COX 1098 00:51:29,937 --> 00:51:32,472 REGRESSION MODELS. 1099 00:51:32,472 --> 00:51:35,576 ENHANCE ACCURACY OF SURVIVAL 1100 00:51:35,576 --> 00:51:38,378 ANALYSIS AND RISK PREDICTION, 1101 00:51:38,378 --> 00:51:40,314 SUPPORTS CUTTING-EDGE ANALYTICS 1102 00:51:40,314 --> 00:51:43,250 IN THE REAL WORLD DATA CONTEXT. 1103 00:51:43,250 --> 00:51:44,017 NEXT SLIDE. 1104 00:51:44,017 --> 00:51:46,887 I THINK THIS MIGHT BE IT. 1105 00:51:46,887 --> 00:51:52,125 NEXT SLIDE. 1106 00:51:52,125 --> 00:51:54,428 OH, OTHER PLANNED PROJECTS. 1107 00:51:54,428 --> 00:51:59,466 WE WANT TO LOOK AT BEING FAIR, 1108 00:51:59,466 --> 00:52:01,768 BOTH SEXES, LOOK AT TESTOSTERONE 1109 00:52:01,768 --> 00:52:04,338 REPLACEMENT THERAPY AND EFFECT 1110 00:52:04,338 --> 00:52:05,806 ON PROSTATE DISORDERS. 1111 00:52:05,806 --> 00:52:08,442 CURRENTLY THERE'S A PAPER WE 1112 00:52:08,442 --> 00:52:08,976 SUBMITTED WHICH IS UNDER 1113 00:52:08,976 --> 00:52:13,513 REVISION IN THE JOURNAL OF 1114 00:52:13,513 --> 00:52:16,183 CLINICAL ENDOCRINOLOGY AND 1115 00:52:16,183 --> 00:52:17,451 METABOLISM. 1116 00:52:17,451 --> 00:52:20,654 TESTOSTERONE SLIGHTLY INCREASED 1117 00:52:20,654 --> 00:52:23,357 RISK OF BENIGN PROSTATIC 1118 00:52:23,357 --> 00:52:24,992 HYPERTROPHY BY 3%, DECREASING 1119 00:52:24,992 --> 00:52:28,095 RISK OF PROSTATE CANCER BY 19%. 1120 00:52:28,095 --> 00:52:30,864 WE DON'T HAVE A BIOLOGIC 1121 00:52:30,864 --> 00:52:35,435 EXPLANATION FOR THAT. 1122 00:52:35,435 --> 00:52:38,705 IT INCREASED BENIGN PROSTATE 1123 00:52:38,705 --> 00:52:41,408 HYPERTROPHY, MINIMIZED WITH 1124 00:52:41,408 --> 00:52:43,610 TOPICAL, LONG-TERM, AVOIDING 1125 00:52:43,610 --> 00:52:49,082 AROMATASE INHIBITOR 1126 00:52:49,082 --> 00:52:49,850 CO-ADMINISTRATION. 1127 00:52:49,850 --> 00:52:51,852 NEXT SLIDE. 1128 00:52:51,852 --> 00:52:54,554 SO, ONGOING AND PLANNED 1129 00:52:54,554 --> 00:52:55,722 PROJECTS, WE'RE LOOKING -- WE 1130 00:52:55,722 --> 00:53:01,395 DRAFTED A PAPER WHICH IS UNDER 1131 00:53:01,395 --> 00:53:03,931 REVISION, ABOUT GLP-1 AND OPTIC 1132 00:53:03,931 --> 00:53:08,335 NEUROPATHY RISK, NOW UNDER 1133 00:53:08,335 --> 00:53:11,071 REVISION AT JAMA OPHTHALMOLOGY 1134 00:53:11,071 --> 00:53:14,274 COMPARED TO SECOND-LINE DIABETES 1135 00:53:14,274 --> 00:53:15,909 MEDICATIONS, GLP-1 RA, INCREASED 1136 00:53:15,909 --> 00:53:20,447 RISK OF OPTIC NEUROPATHY BY 16%. 1137 00:53:20,447 --> 00:53:23,350 AMONG FOUR GLP-1 RAs, ONLY 1138 00:53:23,350 --> 00:53:33,894 SEMAGLUTIDE AND LIRAGLUTIDE WERE 1139 00:53:38,332 --> 00:53:40,067 RINKED TO INCREASED RISK. 1140 00:53:40,067 --> 00:53:43,937 WE HAVE A PAPER ABOUT 1141 00:53:43,937 --> 00:53:46,573 INTEGRATION FRAMEWORK UNDER 1142 00:53:46,573 --> 00:53:53,447 REVIEW, DIABETES MED AND 1143 00:53:53,447 --> 00:53:54,481 PANCREATITIS, PRELIMINARY 1144 00:53:54,481 --> 00:53:56,083 RESULTS INDICATE GLP-1 RA NOT 1145 00:53:56,083 --> 00:53:59,686 ASSOCIATED WITH INCREASED RISK 1146 00:53:59,686 --> 00:54:04,157 OF PANCREATITIS IN CONTRAST TO 1147 00:54:04,157 --> 00:54:07,127 PRIOR EVIDENCE. 1148 00:54:07,127 --> 00:54:09,029 NEXT SLIDE. 1149 00:54:09,029 --> 00:54:17,204 WE HAVE ONGOING AND PLANNED 1150 00:54:17,204 --> 00:54:18,405 PROJECTS, COMPARISONS, BLEEDING 1151 00:54:18,405 --> 00:54:28,749 RISK, IN PREPARATION, A PAPER IN 1152 00:54:28,749 --> 00:54:29,116 PREPARATION. 1153 00:54:29,116 --> 00:54:36,023 PRELIMINARY FINDINGS INDICATE A 1154 00:54:36,023 --> 00:54:36,823 RISK. 1155 00:54:36,823 --> 00:54:37,858 THIS WAS DISSEMINATED BY A 1156 00:54:37,858 --> 00:54:40,427 STUDENT WHO WORKED WITH US FOR A 1157 00:54:40,427 --> 00:54:41,862 PERIOD OF TIME. 1158 00:54:41,862 --> 00:54:45,365 WE ALSO ARE WORKING ON DEEP 1159 00:54:45,365 --> 00:54:47,000 LEARNING, A NEW METHOD, A METHOD 1160 00:54:47,000 --> 00:54:49,169 FOR SURVIVAL ANALYSIS. 1161 00:54:49,169 --> 00:54:51,605 THIS IS IN PREPARATION, IT'S 1162 00:54:51,605 --> 00:54:56,777 PRETTY COOL. 1163 00:54:56,777 --> 00:54:57,744 NEXT SLIDE. 1164 00:54:57,744 --> 00:54:59,246 SUMMARY IMPACT WE MADE 1165 00:54:59,246 --> 00:55:02,649 SUBSTANTIAL CONTRIBUTION TO DRUG 1166 00:55:02,649 --> 00:55:03,784 SAFETY AND COVID-19 RESEARCH, 1167 00:55:03,784 --> 00:55:08,755 USED BIG DATA TO SUPPORT 1168 00:55:08,755 --> 00:55:10,524 CLINICAL DECISION MAKING. 1169 00:55:10,524 --> 00:55:12,526 WE'VE INNOVATED IN METHODOLOGY 1170 00:55:12,526 --> 00:55:17,931 AND DATABASE INFRASTRUCTURE. 1171 00:55:17,931 --> 00:55:19,433 AND ONGOING WORK POSITIONS TEAM 1172 00:55:19,433 --> 00:55:23,036 FOR CONTINUED LEADERSHIP. 1173 00:55:23,036 --> 00:55:24,104 NEXT SLIDE. 1174 00:55:24,104 --> 00:55:28,675 SO I WANT TO THANK 1175 00:55:28,675 --> 00:55:30,110 COLLABORATORS, DATA PARTNERS, 1176 00:55:30,110 --> 00:55:33,613 RESEARCH TEAM, ESPECIALLY 1177 00:55:33,613 --> 00:55:35,348 APPRECIATION TO NLM REVIEWERS. 1178 00:55:35,348 --> 00:55:39,586 SO QUESTIONS AND FEEDBACK ARE 1179 00:55:39,586 --> 00:55:48,762 WELCOME NOW. 1180 00:55:48,762 --> 00:55:51,731 >> GO AHEAD. 1181 00:55:51,731 --> 00:55:54,468 >> HI, CLEM. 1182 00:55:54,468 --> 00:55:56,069 THIS IS BILL HOGAN. 1183 00:55:56,069 --> 00:55:57,304 CURIOUS HOW YOU DECIDE WHICH 1184 00:55:57,304 --> 00:56:02,209 PROBLEMS TO WORK ON, HOW IS THAT 1185 00:56:02,209 --> 00:56:09,583 INFORMED? 1186 00:56:09,583 --> 00:56:09,716 1187 00:56:09,716 --> 00:56:13,553 >> I THINK THERE'S SORT OF 1188 00:56:13,553 --> 00:56:14,521 OPPORTUNITY -- OPPORTUNISTIC, SO 1189 00:56:14,521 --> 00:56:15,455 SOMETHING GETS A LOT OF 1190 00:56:15,455 --> 00:56:20,026 ATTENTION IN THE RESEARCH WORLD, 1191 00:56:20,026 --> 00:56:21,094 WE'LL JUMP ON IT. 1192 00:56:21,094 --> 00:56:25,499 SO IT'S NOT A VERY SCIENTIFIC 1193 00:56:25,499 --> 00:56:26,032 PROCESS. 1194 00:56:26,032 --> 00:56:27,167 DO YOU HAVE ANY THOUGHTS ABOUT 1195 00:56:27,167 --> 00:56:28,635 HOW WE COULD DO IT? 1196 00:56:28,635 --> 00:56:30,604 >> JUST WONDERING IF THERE'S -- 1197 00:56:30,604 --> 00:56:33,106 SINCE THIS KIND OF IS IN THE 1198 00:56:33,106 --> 00:56:35,308 SPACE OF POST-MARKETING 1199 00:56:35,308 --> 00:56:37,644 SURVEILLANCE, IS THERE ANY 1200 00:56:37,644 --> 00:56:38,512 DISCUSSION WITH FDA AND WHAT 1201 00:56:38,512 --> 00:56:42,149 THEY ARE KIND OF WORRIED ABOUT 1202 00:56:42,149 --> 00:56:42,983 OR INTERESTED IN PURSUING? 1203 00:56:42,983 --> 00:56:46,753 >> WE HAVEN'T, BUT IT WOULD BE 1204 00:56:46,753 --> 00:56:50,657 A GOOD IDEA. 1205 00:56:50,657 --> 00:56:53,960 >> AND I'M -- IF I MAY GET 1206 00:56:53,960 --> 00:56:55,629 ANOTHER QUESTION IN, THE UNIFIED 1207 00:56:55,629 --> 00:56:57,130 CODE IN LOINC THAT YOU'RE DOING 1208 00:56:57,130 --> 00:57:00,100 IS THAT RELEVANT TO YOUR OWN 1209 00:57:00,100 --> 00:57:01,434 DATA ANALYSIS OR, YOU KNOW, I 1210 00:57:01,434 --> 00:57:03,837 WAS THINKING IF YOU'RE USING 1211 00:57:03,837 --> 00:57:05,572 MEDICARE CLAIMS, MAYBE THE LAB 1212 00:57:05,572 --> 00:57:06,873 DATA AREN'T THERE, MAYBE THEY 1213 00:57:06,873 --> 00:57:15,081 ARE AND I'M JUST NOT AWARE. 1214 00:57:15,081 --> 00:57:19,586 >> WELL, I'M NOT SURE HOW TO 1215 00:57:19,586 --> 00:57:19,886 ANSWER THAT. 1216 00:57:19,886 --> 00:57:24,491 IT IS -- IT'S USED WIDELY IN THE 1217 00:57:24,491 --> 00:57:26,092 HEALTHCARE SYSTEM, LOINC IS. 1218 00:57:26,092 --> 00:57:30,096 I DON'T THINK MEDICARE ACTUALLY 1219 00:57:30,096 --> 00:57:31,264 ACTIVELY USES IT. 1220 00:57:31,264 --> 00:57:34,634 NOT SO SURE THAT ANSWERED YOUR 1221 00:57:34,634 --> 00:57:38,305 QUESTION. 1222 00:57:38,305 --> 00:57:42,609 >> SO I GUESS I WAS JUST ASKING 1223 00:57:42,609 --> 00:57:46,846 IF IT'S -- IF THE LOINC WORK WAS 1224 00:57:46,846 --> 00:57:49,649 RELEVANT TO YOUR DATA ANALYTICS 1225 00:57:49,649 --> 00:57:55,889 ON DRUGS. 1226 00:57:55,889 --> 00:57:58,892 >> OH. 1227 00:57:58,892 --> 00:58:09,369 I DON'T KNOW, TO BE HONEST. 1228 00:58:16,710 --> 00:58:18,245 >> THIS IS KING JORDAN, GEORGIA 1229 00:58:18,245 --> 00:58:18,445 TECH. 1230 00:58:18,445 --> 00:58:20,814 THANKS FOR YOUR PRESENTATION. 1231 00:58:20,814 --> 00:58:22,282 I'M WONDERING ABOUT THE 1232 00:58:22,282 --> 00:58:25,585 IMPLICATIONS OF YOUR WORK. 1233 00:58:25,585 --> 00:58:28,288 OBVIOUSLY VERY RELEVANT FOR 1234 00:58:28,288 --> 00:58:29,756 POPULATIONS LEFT OUT OF RCTs 1235 00:58:29,756 --> 00:58:32,292 SO COULD HAVE AN IMPACT BOTH ON 1236 00:58:32,292 --> 00:58:36,529 CLINICAL DECISION MAKES, -- 1237 00:58:36,529 --> 00:58:37,964 MAKING BUT ALSO AS POLICY. 1238 00:58:37,964 --> 00:58:39,799 HOW DOES YOUR GROUP WORK TO 1239 00:58:39,799 --> 00:58:40,767 TRANSLATE THESE RESULTS TO THE 1240 00:58:40,767 --> 00:58:41,067 REAL WORLD? 1241 00:58:41,067 --> 00:58:43,069 DO YOU HAVE CONNECTIONS WITH 1242 00:58:43,069 --> 00:58:43,737 POLICYMAKERS? 1243 00:58:43,737 --> 00:58:44,971 DO YOU INTERACT WITH CLINICAL 1244 00:58:44,971 --> 00:58:49,542 DECISION MAKERS? 1245 00:58:49,542 --> 00:58:51,811 I'M CURIOUS ABOUT THAT. 1246 00:58:51,811 --> 00:58:56,583 >> I DON'T THINK WE HAVE DIRECT 1247 00:58:56,583 --> 00:59:00,787 CONNECTIONS, EXCEPT THROUGH 1248 00:59:00,787 --> 00:59:02,055 PUBLIC FACING POLICYMAKERS. 1249 00:59:02,055 --> 00:59:02,956 I WISH WE DID. 1250 00:59:02,956 --> 00:59:13,366 BUT I DON'T THINK WE DO. 1251 00:59:16,803 --> 00:59:19,205 DO YOU HAVE ANY SUGGESTIONS? 1252 00:59:19,205 --> 00:59:20,940 >> NO, I DON'T, UNFORTUNATELY. 1253 00:59:20,940 --> 00:59:23,143 THAT'S ABOVE MY PAY GRADE. 1254 00:59:23,143 --> 00:59:26,246 BUT THAT SEEMS LIKE AN 1255 00:59:26,246 --> 00:59:26,646 OPPORTUNITY HERE. 1256 00:59:26,646 --> 00:59:30,283 YOU KNOW, THE WORK IS CLEARLY 1257 00:59:30,283 --> 00:59:32,385 RELEVANT IN THAT SPACE. 1258 00:59:32,385 --> 00:59:35,088 SO -- AND THE ATTENTION THE 1259 00:59:35,088 --> 00:59:38,558 MENOPAUSE PAPER GOT SUGGESTS THE 1260 00:59:38,558 --> 00:59:44,097 POSSIBILITY TO MAKE A REAL WORLD 1261 00:59:44,097 --> 00:59:46,466 IMPACT.. 1262 00:59:46,466 --> 00:59:48,034 I HAVE MORE QUESTIONS BUT I'LL 1263 00:59:48,034 --> 00:59:52,072 HOLD OFF AND PASS IT. 1264 00:59:52,072 --> 00:59:54,274 >> I CAN'T SEE IF SOMEONE ELSE 1265 00:59:54,274 --> 00:59:55,809 RAISED A HAND. 1266 00:59:55,809 --> 00:59:56,176 IMPRESSIVE WORK. 1267 00:59:56,176 --> 00:59:59,546 THANK YOU SO MUCH FOR SHARING. 1268 00:59:59,546 --> 01:00:02,382 I SEE YOUR OUTSTANDING NUMBER OF 1269 01:00:02,382 --> 01:00:03,416 PUBLICATIONS AND SUCH, WONDERING 1270 01:00:03,416 --> 01:00:07,454 SOMEONE AT YOUR LEVEL OF 1271 01:00:07,454 --> 01:00:10,790 SENIORITY DOING THIS IMPRESSIVE 1272 01:00:10,790 --> 01:00:13,593 WORK HOW DO YOU FOLLOW UP WITH 1273 01:00:13,593 --> 01:00:16,196 THE PREVIEWS, QUESTIONS, WITH 1274 01:00:16,196 --> 01:00:16,830 DIFFERENT ANGLE, COLLABORATIONS, 1275 01:00:16,830 --> 01:00:19,566 WHAT YOU FEEL IS YOUR MENTORING 1276 01:00:19,566 --> 01:00:22,535 ROLE WITHIN NIH AND OUTSIDE NIH, 1277 01:00:22,535 --> 01:00:32,946 YOU KNOW, FOR THE NEW GENERATION 1278 01:00:32,946 --> 01:00:36,349 TO PASS ALL THIS? 1279 01:00:36,349 --> 01:00:38,985 >> CAN YOU RESTATE THE QUESTION? 1280 01:00:38,985 --> 01:00:42,722 >> YES, MY QUESTION WAS ABOUT 1281 01:00:42,722 --> 01:00:44,557 MENTORING AND I MEAN I DON'T 1282 01:00:44,557 --> 01:00:47,761 KNOW IF YOU HEAR ANY PART OF 1283 01:00:47,761 --> 01:00:49,396 WHAT I SAID, I'LL START BY 1284 01:00:49,396 --> 01:00:52,699 SAYING THAT YOU HAVE IMPRESSIVE 1285 01:00:52,699 --> 01:00:56,236 NUMBER OF PUBLICATIONS AND 1286 01:00:56,236 --> 01:00:57,337 DEFINITELY GREAT IMPACT IN OUR 1287 01:00:57,337 --> 01:01:00,607 RESEARCH WORLD AND I WAS 1288 01:01:00,607 --> 01:01:03,743 WONDERING ABOUT HOW DO YOU 1289 01:01:03,743 --> 01:01:04,944 ENVISION YOUR MENTORING, PASSING 1290 01:01:04,944 --> 01:01:06,679 THOSE SKILLS, BOTH AS A MENTOR 1291 01:01:06,679 --> 01:01:14,754 AND ALSO AS A COLLABORATOR. 1292 01:01:14,754 --> 01:01:19,592 >> SO, WE'RE -- I DON'T HAVE A 1293 01:01:19,592 --> 01:01:22,395 REGULAR FLOW OF MENTEES, AT 1294 01:01:22,395 --> 01:01:25,698 LEAST I HAVEN'T, BECAUSE I'M 1295 01:01:25,698 --> 01:01:27,867 REMOTE FROM NIH. 1296 01:01:27,867 --> 01:01:31,070 BUT WE HAVE HAD SOME GOOD 1297 01:01:31,070 --> 01:01:34,340 MENTEES, BECAUSE THERE'S A GUY 1298 01:01:34,340 --> 01:01:37,544 NAMED LUKE BEST WE MIGHT HAVE 1299 01:01:37,544 --> 01:01:40,680 MENTIONED, NOW AT BRIGHAM AND 1300 01:01:40,680 --> 01:01:41,514 WOMEN'S, WHO WE WERE FORTUNATE 1301 01:01:41,514 --> 01:01:49,489 TO SEE SUCH A QUICK STUDY, HE 1302 01:01:49,489 --> 01:01:51,991 ACTUALLY DID THE STUDY, SO WE 1303 01:01:51,991 --> 01:01:53,092 LOVE HAVING GOOD MENTEES. 1304 01:01:53,092 --> 01:01:57,697 AND I GUESS WE'VE BEEN LUCKY TO 1305 01:01:57,697 --> 01:01:58,665 HAVE HAD THEM. 1306 01:01:58,665 --> 01:02:00,800 DOES THAT ANSWER YOUR QUESTION? 1307 01:02:00,800 --> 01:02:10,944 >> YES, THANK YOU. 1308 01:02:10,944 --> 01:02:14,414 >> ANY NON-BOARD OF SCIENTIFIC 1309 01:02:14,414 --> 01:02:17,517 COUNSELOR QUESTIONS WHILE WE'RE 1310 01:02:17,517 --> 01:02:25,658 STILL IN PUBLIC SESSION? 1311 01:02:25,658 --> 01:02:28,795 OR MORE BOARD OF SCIENTIFIC 1312 01:02:28,795 --> 01:02:38,338 COUNSELOR QUESTIONS? 1313 01:02:38,338 --> 01:02:40,240 >> I HAVE ANOTHER QUESTION. 1314 01:02:40,240 --> 01:02:42,675 I NOTICED ON YOUR SLIDES THE CMS 1315 01:02:42,675 --> 01:02:47,347 DATA THAT YOU'RE USING ENDS IN 1316 01:02:47,347 --> 01:02:48,047 2022. 1317 01:02:48,047 --> 01:02:49,349 YOU DID -- OR GOES THROUGH 2022. 1318 01:02:49,349 --> 01:02:53,520 YOU DID A LOT OF WORK ON COVID, 1319 01:02:53,520 --> 01:02:55,421 OBVIOUSLY A MAJOR FOCUS OF YOUR 1320 01:02:55,421 --> 01:02:56,256 GROUP DURING THIS PERIOD. 1321 01:02:56,256 --> 01:02:58,591 WHAT KIND OF CHANGES DO YOU 1322 01:02:58,591 --> 01:03:01,261 ANTICIPATE IN THE RESULTS, THE 1323 01:03:01,261 --> 01:03:04,564 DATA YOU'RE SEEING, AS THAT CMS 1324 01:03:04,564 --> 01:03:10,236 DATA GO FURTHER BEYOND THE 1325 01:03:10,236 --> 01:03:10,503 PANDEMIC? 1326 01:03:10,503 --> 01:03:12,205 >> GOOD QUESTION. 1327 01:03:12,205 --> 01:03:16,376 WELL, THERE WAS CERTAINLY THE 1328 01:03:16,376 --> 01:03:20,647 SPIKE FROM COVID WILL DIMINISH. 1329 01:03:20,647 --> 01:03:25,285 BUT OTHERWISE, I MEAN, IT'S A 1330 01:03:25,285 --> 01:03:26,553 PRETTY STABLE POPULATION, 1331 01:03:26,553 --> 01:03:27,220 MEDICARE POPULATION. 1332 01:03:27,220 --> 01:03:28,922 THEY JUST GET OLDER. 1333 01:03:28,922 --> 01:03:31,624 AND BUT I DON'T THINK WE SEE 1334 01:03:31,624 --> 01:03:34,460 ANYTHING DRAMATIC AFTER THE -- 1335 01:03:34,460 --> 01:03:39,632 EXCEPT POST-COVID. 1336 01:03:39,632 --> 01:03:40,233 I MEAN, FORGETTING ABOUT COVID 1337 01:03:40,233 --> 01:03:44,304 FOR A MOMENT. 1338 01:03:44,304 --> 01:03:46,639 >> SIMILARLY I NOTICE YOU'RE 1339 01:03:46,639 --> 01:03:47,440 WORKING ON GLPs. 1340 01:03:47,440 --> 01:03:49,976 THINKING AGAIN IF THE DATA GO 1341 01:03:49,976 --> 01:03:53,880 THROUGH 2022, MOST OF THE ACTION 1342 01:03:53,880 --> 01:03:55,014 OBVIOUSLY WITH GLPs HAS 1343 01:03:55,014 --> 01:03:56,783 HAPPENED VERY, VERY RECENTLY. 1344 01:03:56,783 --> 01:03:58,284 WHAT ARE YOU THINKING ABOUT IN 1345 01:03:58,284 --> 01:04:02,388 THAT SPACE AS THEY START TO ROLL 1346 01:04:02,388 --> 01:04:04,757 OUT MORE UPDATED DATA? 1347 01:04:04,757 --> 01:04:06,059 WHEN WERE GLPs FIRST APPROVED 1348 01:04:06,059 --> 01:04:07,760 AND PUT ON THE MARKET? 1349 01:04:07,760 --> 01:04:10,496 >> I DON'T KNOW EITHER, THE 1350 01:04:10,496 --> 01:04:11,197 LAST COUPLE YEARS. 1351 01:04:11,197 --> 01:04:12,699 I THINK THAT'S A VERY 1352 01:04:12,699 --> 01:04:14,100 INTERESTING THING TO LOOK AT. 1353 01:04:14,100 --> 01:04:17,070 I'M SURE IT WILL HAVE SOME 1354 01:04:17,070 --> 01:04:17,303 EFFECT. 1355 01:04:17,303 --> 01:04:19,505 I WOULD BET IT WOULD HAVE SOME 1356 01:04:19,505 --> 01:04:24,077 EFFECT, BECAUSE THEY ARE BEING 1357 01:04:24,077 --> 01:04:25,378 USED WIDELY, PRETTY EFFECTIVE 1358 01:04:25,378 --> 01:04:28,348 DRUGS, SO WE'LL LOOK AT THAT TO 1359 01:04:28,348 --> 01:04:31,818 SEE IF WE CAN SEE A BUMP IN 1360 01:04:31,818 --> 01:04:33,586 ANYTHING WITH RELATIONSHIP TO 1361 01:04:33,586 --> 01:04:37,223 AVAILABILITY OF THE GLP DRUGS. 1362 01:04:37,223 --> 01:04:40,560 >> CLEM, I HAVE A QUESTION WHICH 1363 01:04:40,560 --> 01:04:41,194 MERGES TWO QUESTIONS. 1364 01:04:41,194 --> 01:04:43,229 ONE OF WHICH WAS THE EARLIER 1365 01:04:43,229 --> 01:04:45,598 QUESTION, HOW DO YOU CHOOSE 1366 01:04:45,598 --> 01:04:49,102 AREAS TO INVESTIGATE, AS WELL AS 1367 01:04:49,102 --> 01:04:50,336 YOU DISCUSSED SOME ONGOING AND 1368 01:04:50,336 --> 01:04:52,805 PLANNED PROJECTS BUT I'D BE 1369 01:04:52,805 --> 01:05:00,146 CURIOUS WHAT YOU SEE AS 1370 01:05:00,146 --> 01:05:10,690 POTENTIALLY EMERGING BEYOND GHB 1371 01:05:22,702 --> 01:05:23,069 GLP. 1372 01:05:23,069 --> 01:05:26,506 >> IT'S NEW, SO THAT DREW US TO 1373 01:05:26,506 --> 01:05:27,106 IT, PRETTY STRONGLY. 1374 01:05:27,106 --> 01:05:31,944 I DON'T KNOW IF THERE'S 1375 01:05:31,944 --> 01:05:34,213 COMPARABLE -- ANYTHING 1376 01:05:34,213 --> 01:05:36,149 COMPARABLE BESIDES THAT RIGHT 1377 01:05:36,149 --> 01:05:42,155 NOW IN LABORATORIES, ET CETERA. 1378 01:05:42,155 --> 01:05:43,856 IF I HAD AN HOUR I MIGHT BE ABLE 1379 01:05:43,856 --> 01:05:48,561 TO THINK OF SOMETHING. 1380 01:05:48,561 --> 01:05:57,837 >> FAIR ENOUGH. 1381 01:05:57,837 --> 01:05:58,805 >> OTHER QUESTIONS? 1382 01:05:58,805 --> 01:06:00,707 >> I HAVE A SIMILAR QUESTION 1383 01:06:00,707 --> 01:06:01,474 ABOUT AREAS OF FOCUS. 1384 01:06:01,474 --> 01:06:03,743 IT CAUGHT MY ATTENTION THAT YOU 1385 01:06:03,743 --> 01:06:04,811 MENTIONED ONE OF YOUR -- ONE OF 1386 01:06:04,811 --> 01:06:07,747 THE THINGS THAT INTERESTS YOU IS 1387 01:06:07,747 --> 01:06:10,083 CONTROVERSIAL CLAIMS ABOUT THE 1388 01:06:10,083 --> 01:06:11,984 EFFICACY OR SAFETY OF DRUGS N 1389 01:06:11,984 --> 01:06:15,822 COVID THERE'S OF COURSE ALL 1390 01:06:15,822 --> 01:06:17,857 KINDS OF CONTROVERSY SWIRLING 1391 01:06:17,857 --> 01:06:19,792 OVER IVERMECTIN, DO YOU HAVE ANY 1392 01:06:19,792 --> 01:06:27,366 PLANS TO GO INTO THAT SPACE W 1393 01:06:27,366 --> 01:06:29,736 ITH RESPECT TO COVID? 1394 01:06:29,736 --> 01:06:30,870 >> MANY THINGS WITH RESPECT TO 1395 01:06:30,870 --> 01:06:31,537 COVID. 1396 01:06:31,537 --> 01:06:32,672 THE DRUG YOU MENTIONED, I DON'T 1397 01:06:32,672 --> 01:06:36,142 THINKS WE HAVE ACCESS TO DATA. 1398 01:06:36,142 --> 01:06:39,312 ISN'T THAT AN ANIMAL DRUG? 1399 01:06:39,312 --> 01:06:39,779 OR THAT'S -- 1400 01:06:39,779 --> 01:06:40,513 >> IT IS. 1401 01:06:40,513 --> 01:06:43,816 IT'S USED IN HUMANS TOO. 1402 01:06:43,816 --> 01:06:45,351 IT'S BEEN PRESCRIBED BUT OFF 1403 01:06:45,351 --> 01:06:45,585 LABEL. 1404 01:06:45,585 --> 01:06:47,253 IT'S NOT APPROVED AS OF YET SO I 1405 01:06:47,253 --> 01:06:49,789 DON'T KNOW IF THAT DATA WOULD 1406 01:06:49,789 --> 01:06:51,023 EVEN COME THROUGH IN CMS. 1407 01:06:51,023 --> 01:06:54,193 >> WE'LL GET DATA ON IT BUT IT'S 1408 01:06:54,193 --> 01:06:55,328 AN INTERESTING -- MY 1409 01:06:55,328 --> 01:06:56,829 UNDERSTANDING THAT DOESN'T DO 1410 01:06:56,829 --> 01:06:58,531 ANY GOOD ANYWAY. 1411 01:06:58,531 --> 01:07:03,169 I MEAN, IT WAS JUST AN IDEA, 1412 01:07:03,169 --> 01:07:05,505 SOME PEOPLE ON THE EDGE, BUT I 1413 01:07:05,505 --> 01:07:06,272 MAY BE WRONG. 1414 01:07:06,272 --> 01:07:08,775 BUT I DON'T THINK IT'S REALLY 1415 01:07:08,775 --> 01:07:14,180 USEFUL DRUG FOR THE DISEASE. 1416 01:07:14,180 --> 01:07:15,081 MY RECOLLECTION. 1417 01:07:15,081 --> 01:07:17,550 SO BUT WE COULD LOOK AT THAT. 1418 01:07:17,550 --> 01:07:18,785 BUT I DON'T KNOW WHERE WE'RE 1419 01:07:18,785 --> 01:07:20,419 GOING TO GET DRUG DATA ON 1420 01:07:20,419 --> 01:07:21,854 IVERMECTIN BECAUSE I THINK IT'S 1421 01:07:21,854 --> 01:07:22,688 A VETERINARY DRUG. 1422 01:07:22,688 --> 01:07:24,924 WE MIGHT BE ABLE TO GET IT. 1423 01:07:24,924 --> 01:07:27,460 I DON'T KNOW. 1424 01:07:27,460 --> 01:07:31,497 IS THAT RIGHT? 1425 01:07:31,497 --> 01:07:33,499 >> IT'S A HUMAN DRUG TOO. 1426 01:07:33,499 --> 01:07:34,667 IT'S BEEN PRESCRIBED, BUT OFF 1427 01:07:34,667 --> 01:07:35,968 LABEL FOR COVID. 1428 01:07:35,968 --> 01:07:37,503 SO SINCE I DON'T KNOW MUCH ABOUT 1429 01:07:37,503 --> 01:07:40,640 THE CMS SYSTEM AND HOW THAT 1430 01:07:40,640 --> 01:07:41,307 WORKS, MEDICAID AND MEDICARE, 1431 01:07:41,307 --> 01:07:47,914 YOU MIGHT NOT BE ABLE TO GET IN. 1432 01:07:47,914 --> 01:07:49,115 IN GENERAL BEYOND IVERMECTIN 1433 01:07:49,115 --> 01:07:52,084 SPECIFICALLY IS A FOCUS AREA IS 1434 01:07:52,084 --> 01:07:54,220 CONTROVERSIAL DRUGS WHEN COVID 1435 01:07:54,220 --> 01:07:56,956 WOULD BE RIPE FOR THE 1436 01:07:56,956 --> 01:07:57,824 RETROSPECTIVE STUDIES YOU'RE 1437 01:07:57,824 --> 01:07:58,024 DOING. 1438 01:07:58,024 --> 01:07:59,358 >> I THINK THAT'S A GREAT IDEA. 1439 01:07:59,358 --> 01:08:09,802 WE'LL TAKE A LOOK AT IT. 1440 01:08:10,670 --> 01:08:20,780 SEAN, DO YOU HAVE ANY COMMENTS? 1441 01:08:20,780 --> 01:08:23,182 >> YOU'RE WELCOME TO BREAK INTO 1442 01:08:23,182 --> 01:08:27,486 YOUR CLOSED SESSION EARLY IF YOU 1443 01:08:27,486 --> 01:08:27,720 WISH. 1444 01:08:27,720 --> 01:08:28,521 >> ALL RIGHT. 1445 01:08:28,521 --> 01:08:32,091 WHY DON'T WE GO AHEAD AND DO 1446 01:08:32,091 --> 01:08:32,525 THAT. 1447 01:08:32,525 --> 01:08:36,028 SO, AT THIS POINT, THE BOARD OF 1448 01:08:36,028 --> 01:08:38,264 SCIENTIFIC COUNSELORS WILL 1449 01:08:38,264 --> 01:08:40,499 CONVENE A CLOSED SESSION WITH 1450 01:08:40,499 --> 01:08:43,236 JUST CLEM FOR FURTHER QUESTIONS 1451 01:08:43,236 --> 01:08:45,238 AND ANSWERS. 1452 01:08:45,238 --> 01:08:47,573 AFTER WHICH WE WILL -- AFTER 15 1453 01:08:47,573 --> 01:08:51,177 MINUTES WE'LL BREAK FOR A CLOSED 1454 01:08:51,177 --> 01:08:53,246 SESSION OF ASSESSMENT AND WE'LL 1455 01:08:53,246 --> 01:08:56,048 RESUME THE PUBLIC SESSION AT 1456 01:08:56,048 --> 01:08:59,342 11:20:00 A.M. EASTERN TIME. 1457 01:08:59,342 --> 01:09:00,443 ALL RIGHT. 1458 01:09:00,443 --> 01:09:03,046 SO, WE'RE OPENING UP THE 1459 01:09:03,046 --> 01:09:04,013 RECORDING, AND THE PUBLIC 1460 01:09:04,013 --> 01:09:06,582 SESSION FOR THE NEXT PART OF THE 1461 01:09:06,582 --> 01:09:07,216 MEETING OF THE NATIONAL LIBRARY 1462 01:09:07,216 --> 01:09:12,321 OF MEDICINE BOARD OF SCIENTIFIC 1463 01:09:12,321 --> 01:09:13,890 COUNSELORS, BROADCAST AS WELL, 1464 01:09:13,890 --> 01:09:20,029 AND I WILL WELCOME LAUREN PORTER 1465 01:09:20,029 --> 01:09:20,963 TO GIVE HER PRESENTATION. 1466 01:09:20,963 --> 01:09:21,464 >> GOOD MORNING. 1467 01:09:21,464 --> 01:09:26,502 I WANT TO TALK ABOUT 1468 01:09:26,502 --> 01:09:30,840 UNDERSTANDING ALTERNATIVE 1469 01:09:30,840 --> 01:09:32,275 PROTEIN CONFIRMATION, AND WE'LL 1470 01:09:32,275 --> 01:09:33,676 TALK ABOUT EXPERIMENT ALSO. 1471 01:09:33,676 --> 01:09:39,515 I'M SURE ALL OF YOU KNOW THAT 1472 01:09:39,515 --> 01:09:41,818 ALPHAFOLD HAS MADE A HUGE IMPACT 1473 01:09:41,818 --> 01:09:43,686 ON STRUCTURAL BIOLOGY, 1474 01:09:43,686 --> 01:09:45,321 PARTICULARLY THE PREDICTION 1475 01:09:45,321 --> 01:09:49,392 OFTEN VERY ACCURATE OF THREE 1476 01:09:49,392 --> 01:09:50,359 DIMENSIONAL STRUCTURES OF 1477 01:09:50,359 --> 01:09:51,994 PROTEIN SEQUENCES. 1478 01:09:51,994 --> 01:09:54,630 SO JUST A MAPPING FROM SEQUENCE 1479 01:09:54,630 --> 01:09:56,299 TO STRUCTURE THAT HAPPENS VERY 1480 01:09:56,299 --> 01:09:58,334 QUICKLY AND HAS MADE SUCH AN 1481 01:09:58,334 --> 01:09:59,635 IMPACT, SO ACCURATE, IT'S BEEN 1482 01:09:59,635 --> 01:10:05,074 USED IN SO MANY CASES THAT IT 1483 01:10:05,074 --> 01:10:06,409 WAS AWARDED THE NOBEL PRIZE A 1484 01:10:06,409 --> 01:10:07,143 FEW MONTHS AGO. 1485 01:10:07,143 --> 01:10:09,979 WITH THAT COMES A QUESTION, 1486 01:10:09,979 --> 01:10:11,747 WHAT'S NEXT FOR PROTEIN 1487 01:10:11,747 --> 01:10:14,150 STRUCTURE? 1488 01:10:14,150 --> 01:10:19,288 AND AS GREG BOWMAN WROTE IN AN 1489 01:10:19,288 --> 01:10:22,325 OPINION PIECE, PREDICTING 1490 01:10:22,325 --> 01:10:23,526 ALTERNATIVE CONFIRMATION IS NEW 1491 01:10:23,526 --> 01:10:23,793 FRONTIER. 1492 01:10:23,793 --> 01:10:26,395 THIS WAS SUPPOSED TO BE A VIDEO 1493 01:10:26,395 --> 01:10:34,337 TO SHOW YOU OUT WARD AND INWARD 1494 01:10:34,337 --> 01:10:39,675 FACING TRANSLATIONS OF THE MONO 1495 01:10:39,675 --> 01:10:41,110 CARBO LATE TRANSPORTER. 1496 01:10:41,110 --> 01:10:45,014 THIS IS SOMETHING THAT ALPHAFOLD 1497 01:10:45,014 --> 01:10:50,286 CAN PREDICT EASILY BUT 1498 01:10:50,286 --> 01:10:51,454 PREDICTING ALTERNATIVE 1499 01:10:51,454 --> 01:10:52,155 CONFIRMATION WHICH IS HERE IN 1500 01:10:52,155 --> 01:10:56,192 PINK IS A HARDER THING TO DO. 1501 01:10:56,192 --> 01:10:58,728 AND SO THIS HAS BECOME A FOCUS 1502 01:10:58,728 --> 01:10:59,595 OF PROTEIN STRUCTURE PREDICTION. 1503 01:10:59,595 --> 01:11:05,501 YOU'LL SEE A LOT IN MY TALK WHEN 1504 01:11:05,501 --> 01:11:08,070 I TALK ABOUT DOMINANT 1505 01:11:08,070 --> 01:11:09,872 CONFIRMATIONS THAT ALPHAFOLD CAN 1506 01:11:09,872 --> 01:11:10,540 PREDICT EASILY. 1507 01:11:10,540 --> 01:11:12,175 ALTERNATIVE IN PINK ARE HARDER 1508 01:11:12,175 --> 01:11:12,542 TO PREDICT. 1509 01:11:12,542 --> 01:11:18,214 SO THIS IS AN EXAMPLE MONO 1510 01:11:18,214 --> 01:11:19,749 CARBOXYLATE TRANSPORTER IS AN 1511 01:11:19,749 --> 01:11:24,387 EXAMPLE OF A RIGID BODY MOTION 1512 01:11:24,387 --> 01:11:26,088 BETWEEN TWO DOMAINS, THERE ARE 1513 01:11:26,088 --> 01:11:27,890 OTHER CHANGES YOU'RE FAMILIAR 1514 01:11:27,890 --> 01:11:34,063 WITH SUCH AS SUBDOMAIN 1515 01:11:34,063 --> 01:11:35,097 REORIENTATIONS, CRYPTIC POCKETS 1516 01:11:35,097 --> 01:11:37,033 IMPORTANT FOR DRUG DESIGN. 1517 01:11:37,033 --> 01:11:39,302 MY LAB LOOKS AT AN EXTREME 1518 01:11:39,302 --> 01:11:40,536 VERSION OF CONFIRMATIONAL CHANGE 1519 01:11:40,536 --> 01:11:42,305 KNOWN AS FOLD SWITCHING. 1520 01:11:42,305 --> 01:11:44,607 WHAT IS FOLD SWITCHING? 1521 01:11:44,607 --> 01:11:51,314 IT IS A DRAMATIC REMODELING OF 1522 01:11:51,314 --> 01:11:53,716 SECONDARY AND TERTIARY STRUCTURE 1523 01:11:53,716 --> 01:11:55,618 THAT LEADS TO CHANGES IN 1524 01:11:55,618 --> 01:12:03,092 FUNCTION AND REGULATION, THIS IS 1525 01:12:03,092 --> 01:12:04,694 RFH, N-TERMINAL DOMAIN MAINTAINS 1526 01:12:04,694 --> 01:12:07,697 THE STRUCTURE BUT C-TERMINAL 1527 01:12:07,697 --> 01:12:09,932 DOMAIN, NOT A PHYSICAL PATHWAY, 1528 01:12:09,932 --> 01:12:14,770 JUST A PATHWAY -- A LINEAR 1529 01:12:14,770 --> 01:12:15,838 PATHWAY BETWEEN TWO 1530 01:12:15,838 --> 01:12:21,611 CONFIRMATION, TWO VERY DIFFERENT 1531 01:12:21,611 --> 01:12:24,614 CONFIRMS, BINDING A SUBUNIT OF 1532 01:12:24,614 --> 01:12:26,616 THE RIBOSOME, ALPHA HELICAL FORM 1533 01:12:26,616 --> 01:12:28,818 IS IMPORTANT FOR SPECIFIC BIND 1534 01:12:28,818 --> 01:12:34,090 TOCK A DNA SITE CALLED OX. 1535 01:12:34,090 --> 01:12:36,792 BASED ON NOTATION THE BETA SHEET 1536 01:12:36,792 --> 01:12:39,895 CONFIRMATION IS EASY ONE TO 1537 01:12:39,895 --> 01:12:41,631 PREDICT, ALPHA HELICAL 1538 01:12:41,631 --> 01:12:43,633 ALTERNATIVE IS MORE DIFFICULT. 1539 01:12:43,633 --> 01:12:46,569 IT TURNS OUT THIS IS ONE EXAMPLE 1540 01:12:46,569 --> 01:12:49,071 OF 100 KNOWN FOLD SWITCHING 1541 01:12:49,071 --> 01:12:50,072 PROTEINS, BUT THERE ARE MANY 1542 01:12:50,072 --> 01:12:52,108 THAT SEEM TO BE INVOLVED IN 1543 01:12:52,108 --> 01:12:57,213 HUMAN DISEASE, SO HERE ARE A FEW 1544 01:12:57,213 --> 01:12:59,148 EXAMPLES, TUBERCULOSIS, CANCER, 1545 01:12:59,148 --> 01:12:59,749 ALZHEIMER'S, GASTROENTERITIS, 1546 01:12:59,749 --> 01:13:00,950 EVEN A COVID-19 PROTEIN THAT'S 1547 01:13:00,950 --> 01:13:04,453 BEEN FOUND TO SWITCH FOLDS. 1548 01:13:04,453 --> 01:13:06,889 NOT ONLY THAT, BUT MINOR 1549 01:13:06,889 --> 01:13:08,691 SEQUENCE MODIFICATIONS CAN ALSO 1550 01:13:08,691 --> 01:13:10,226 CAUSE MAJOR STRUCTURAL CHANGES. 1551 01:13:10,226 --> 01:13:15,264 SO, THIS IS A HUMAN CANCER 1552 01:13:15,264 --> 01:13:17,333 PROTEIN BCCIP, TWO ISOFORMS, 1553 01:13:17,333 --> 01:13:20,236 ALPHA AND BETA, SO THESE TWO 1554 01:13:20,236 --> 01:13:22,505 LINES DOWN HERE IN NAVY BLUE 1555 01:13:22,505 --> 01:13:25,007 REPRESENT 80% OF SEQUENCE, 1556 01:13:25,007 --> 01:13:27,710 COMPLETELY IDENTICAL. 1557 01:13:27,710 --> 01:13:30,646 SO BBPIP ALPHA AND BETA 1558 01:13:30,646 --> 01:13:33,649 SEQUENCES ARE IDENTICAL EXCEPT 1559 01:13:33,649 --> 01:13:36,385 FOR EXON SUBSTITUTION AND THIS 1560 01:13:36,385 --> 01:13:38,254 ONE SUBSTITUTION COMPLETELY 1561 01:13:38,254 --> 01:13:40,389 CHANGES THE STRUCTURES OF THE 1562 01:13:40,389 --> 01:13:41,057 PROTEINS. 1563 01:13:41,057 --> 01:13:42,625 RELATIVE TO EACH OTHER, IF YOU 1564 01:13:42,625 --> 01:13:45,861 ALIGN THEM THEY ARE 13 ANGSTROMS 1565 01:13:45,861 --> 01:13:46,595 DIFFERENT. 1566 01:13:46,595 --> 01:13:48,698 THEY HAVE DIFFERENT FUNCTIONS 1567 01:13:48,698 --> 01:13:50,333 THEMSELVES, SO BECAUSE THERE ARE 1568 01:13:50,333 --> 01:13:54,370 DIFFERENT STRUCTURES ALLOW THEM 1569 01:13:54,370 --> 01:13:55,871 TO HAVE THESE FOLDS. 1570 01:13:55,871 --> 01:13:57,306 OR THESE DIFFERENT FUNCTIONS. 1571 01:13:57,306 --> 01:13:58,974 THE QUESTIONS MY LAB HAS BEEN 1572 01:13:58,974 --> 01:14:01,777 INTERESTED IN ASKING ARE HOW DO 1573 01:14:01,777 --> 01:14:04,547 YOU MINOR CHANGES IN SEQUENCE OR 1574 01:14:04,547 --> 01:14:06,382 ENVIRONMENT CAUSE PROTEINS TO 1575 01:14:06,382 --> 01:14:08,050 SWITCH FOLDS AND FUNCTION? 1576 01:14:08,050 --> 01:14:09,885 THERE'S MANY REASONS WHY THIS 1577 01:14:09,885 --> 01:14:10,519 QUESTION IS SIGNIFICANT, FIRST 1578 01:14:10,519 --> 01:14:13,489 OF ALL THIS IS A REALLY AS I 1579 01:14:13,489 --> 01:14:16,992 SAID EXTREME EXAMPLE OF PROTEIN 1580 01:14:16,992 --> 01:14:17,660 DYNAMICS, CONFIRMATIONAL CHANGE, 1581 01:14:17,660 --> 01:14:19,528 COULD BE AS WE UNDERSTAND IT 1582 01:14:19,528 --> 01:14:21,063 BETTER WE CAN UNDERSTAND OTHER 1583 01:14:21,063 --> 01:14:22,898 SORTS OF CON FIRMATIONAL CHANGES 1584 01:14:22,898 --> 01:14:25,968 AS I'LL GIVE AN EXAMPLE LATER IN 1585 01:14:25,968 --> 01:14:26,635 THE TALK. 1586 01:14:26,635 --> 01:14:31,640 IT COULD LEAD TO DISCOVERY OF 1587 01:14:31,640 --> 01:14:32,842 BIOLOGICAL PATHWAYS, BECAUSE 1588 01:14:32,842 --> 01:14:34,677 THERE ARE FOLD SWITCHING 1589 01:14:34,677 --> 01:14:42,084 PROTEINS KNOWN TO BE INTEGRAL 1590 01:14:42,084 --> 01:14:43,886 INTO BIOLOGICAL PATH WAYS, 1591 01:14:43,886 --> 01:14:45,020 KILLING THE CLOCK. 1592 01:14:45,020 --> 01:14:46,021 MUTATIONS CAN LEAD TO 1593 01:14:46,021 --> 01:14:47,156 ALTERNATIVE FOLDED STATES. 1594 01:14:47,156 --> 01:14:50,359 I JUST SHOWED THAT A PRETTY 1595 01:14:50,359 --> 01:14:51,761 SMALL AXON SUBSTITUTION CHANGES 1596 01:14:51,761 --> 01:14:56,198 A PROTEIN'S FOLD AND FUNCTION. 1597 01:14:56,198 --> 01:14:56,866 ALSO SINGLE NUCLEOTIDE 1598 01:14:56,866 --> 01:14:57,633 POLYMORPHISMS CAN CHANGE PROTEIN 1599 01:14:57,633 --> 01:14:58,968 FOLD AND FUNCTION. 1600 01:14:58,968 --> 01:15:06,542 THERE'S A PROTEIN INVOLVED IN 1601 01:15:06,542 --> 01:15:09,612 NON-HODGKIN LYMPHOMA MEF 2B, SNP 1602 01:15:09,612 --> 01:15:10,913 CAUSES A MUTATION THAT SWITCHING 1603 01:15:10,913 --> 01:15:13,449 C-TERMINAL HELIX TO BETA SHEET, 1604 01:15:13,449 --> 01:15:18,554 BELIEVED TO INACTIVATE THE 1605 01:15:18,554 --> 01:15:23,058 PROTEIN. 1606 01:15:23,058 --> 01:15:26,662 THESE ARE THE QUESTIONS MY LAB 1607 01:15:26,662 --> 01:15:27,797 IS INTERESTED IN ASKING BUT 1608 01:15:27,797 --> 01:15:29,832 THERE ARE BARRIERS TO PROGRESS. 1609 01:15:29,832 --> 01:15:31,801 AS I MENTIONED, THERE ARE ONLY A 1610 01:15:31,801 --> 01:15:34,103 FEW EXAMPLES OF KNOWN FOLD 1611 01:15:34,103 --> 01:15:36,639 SWITCHING PROTEINS, SO THERE ARE 1612 01:15:36,639 --> 01:15:37,506 100 TO DATE. 1613 01:15:37,506 --> 01:15:42,344 SECOND, IN SPITE OF THE AMAZING 1614 01:15:42,344 --> 01:15:43,245 TECHNICAL ADVANCE THE ALPHAFOLD 1615 01:15:43,245 --> 01:15:44,880 TEAM MADE FOLD SWITCHING IS 1616 01:15:44,880 --> 01:15:47,750 DIFFICULT TO PREDICT, UNTIL 1617 01:15:47,750 --> 01:15:49,351 REASONLY NO KNOWN MUTATIONAL 1618 01:15:49,351 --> 01:15:52,421 PATHWAYS WERE KNOWN TO SWITCH 1619 01:15:52,421 --> 01:15:53,389 FOLDS, AND THE FIRST WAS 1620 01:15:53,389 --> 01:15:54,623 DISCOVERED, I CAN'T DISCUSS IN 1621 01:15:54,623 --> 01:15:58,427 THIS TALK BUT CAN SHOW YOU LATER 1622 01:15:58,427 --> 01:15:59,762 IF YOU HAVE QUESTIONS. 1623 01:15:59,762 --> 01:16:01,997 THERE AREN'T RELIABLE METHODS TO 1624 01:16:01,997 --> 01:16:03,265 PREDICT FOLD SWITCHING. 1625 01:16:03,265 --> 01:16:05,134 OUR LAB AND OTHERS MADE SOME 1626 01:16:05,134 --> 01:16:06,502 PROGRESS WHICH I'LL SHOW YOU BUT 1627 01:16:06,502 --> 01:16:09,505 THERE'S STILL A LOT OF ROOM TO 1628 01:16:09,505 --> 01:16:09,705 GROW. 1629 01:16:09,705 --> 01:16:11,173 I THINK PART OF THE REASON FOLD 1630 01:16:11,173 --> 01:16:12,708 SWITCHING IS UNDERREPRESENTED IS 1631 01:16:12,708 --> 01:16:18,047 THAT FOLD SWITCHING PROTEINS ARE 1632 01:16:18,047 --> 01:16:18,647 DIFFICULT TO HANDLE 1633 01:16:18,647 --> 01:16:19,014 EXPERIMENTALLY. 1634 01:16:19,014 --> 01:16:25,287 THIS IS FROM A DAY DOING 1635 01:16:25,287 --> 01:16:26,489 EXPERIMENTS, THIS WHITE STUFF 1636 01:16:26,489 --> 01:16:28,691 WHICH OUR SYSTEM IS PRONE TO 1637 01:16:28,691 --> 01:16:29,725 DOING, WHICH MEANS WE DIDN'T 1638 01:16:29,725 --> 01:16:30,960 COLLECT DATA THAT DAY BECAUSE 1639 01:16:30,960 --> 01:16:33,362 THE SAMPLE WASN'T STABLE ENOUGH 1640 01:16:33,362 --> 01:16:35,030 TO COLLECT IT. 1641 01:16:35,030 --> 01:16:39,768 THIS IS A PERENNIAL CHALLENGE. 1642 01:16:39,768 --> 01:16:48,210 WE'VE BEEN ABLE TO MAKE A BIG 1643 01:16:48,210 --> 01:16:51,247 ADVANCE ON THAT RECENTLY. 1644 01:16:51,247 --> 01:16:56,719 I T TURNS ON YOU FOLD SWITCHING 1645 01:16:56,719 --> 01:16:59,255 IS DIFFICULT TO PREDICT. 1646 01:16:59,255 --> 01:17:03,526 WE HAVE AN X-AXIS PREDICTION 1647 01:17:03,526 --> 01:17:06,629 ACCURACY OF DOMINANT, Y-AXIS 1648 01:17:06,629 --> 01:17:07,229 ALTERNATIVES. 1649 01:17:07,229 --> 01:17:08,364 HERE WE'RE QUANTIFYING ACCURACY 1650 01:17:08,364 --> 01:17:12,635 BY TM SCORE, SAME IF WE WERE 1651 01:17:12,635 --> 01:17:13,702 USING RMSD. 1652 01:17:13,702 --> 01:17:14,904 THIS DOTTED LINE IS IDENTITY 1653 01:17:14,904 --> 01:17:15,271 LINE. 1654 01:17:15,271 --> 01:17:21,377 I WANT YOU TO SEE THAT 94% OF 1655 01:17:21,377 --> 01:17:22,811 PREDICTIONS FALL BELOW THE 1656 01:17:22,811 --> 01:17:24,713 IDENTITY LINE, THAT IS THEY ARE 1657 01:17:24,713 --> 01:17:29,485 SKEWED TOWARD PREDICTING THE 1658 01:17:29,485 --> 01:17:31,787 DOMINANT CONFIRMATION. 1659 01:17:31,787 --> 01:17:33,155 ALTERNATIVE USING VANILLA MODEL 1660 01:17:33,155 --> 01:17:35,057 IS HARDLY PREDICTED AT ALL. 1661 01:17:35,057 --> 01:17:41,764 SO NOW A LOT OF METHODS FOR 1662 01:17:41,764 --> 01:17:44,133 IMPROVING ALPHAFOLD HAVE BEEN 1663 01:17:44,133 --> 01:17:44,400 DEVELOPED. 1664 01:17:44,400 --> 01:17:47,369 THIS CAME FROM TESTING. 1665 01:17:47,369 --> 01:17:50,272 WE GOT A LITTLE BIT OF 1666 01:17:50,272 --> 01:17:55,711 IMPROVEMENT, THESE ARE VANILLA 1667 01:17:55,711 --> 01:17:56,712 IMPLEMENTATIONS. 1668 01:17:56,712 --> 01:17:58,514 FOLD SWITCHING CAN BE TRIGGERED 1669 01:17:58,514 --> 01:18:00,816 BY BINDING ANOTHER PROTEIN OR 1670 01:18:00,816 --> 01:18:02,251 BINDING A DIFFERENT BIOMOLECULE. 1671 01:18:02,251 --> 01:18:05,955 WE TESTED IT ALSO ON ALPHAFOLD2 1672 01:18:05,955 --> 01:18:08,557 MULTIMER, AND 3, AND YOU CAN SEE 1673 01:18:08,557 --> 01:18:10,292 NEITHER DID VERY WELL EITHER 1674 01:18:10,292 --> 01:18:11,327 EVEN WHEN MODELING EVERY 1675 01:18:11,327 --> 01:18:12,928 INTERACTION WE COULD. 1676 01:18:12,928 --> 01:18:16,899 FURTHERMORE THERE WERE ENHANCED 1677 01:18:16,899 --> 01:18:19,868 SAMPLING METHODS THAT WERE MEANT 1678 01:18:19,868 --> 01:18:21,870 TO TIP ALPHAFOLD TO PREDICT ONE 1679 01:18:21,870 --> 01:18:23,172 CONFIRMATION OR ANOTHER AND YOU 1680 01:18:23,172 --> 01:18:30,646 CAN SEE THOSE DID NOT FARE VERY 1681 01:18:30,646 --> 01:18:32,748 WELL EITHER. 1682 01:18:32,748 --> 01:18:35,084 300,000, ALPHAFOLD HAS 35% 1683 01:18:35,084 --> 01:18:37,886 SUCCESS RATE WHICH IS MODEST, 1684 01:18:37,886 --> 01:18:38,454 AND ESPECIALLY CONCERNING 1685 01:18:38,454 --> 01:18:41,056 BECAUSE ALL OF THE STRUCTURES 1686 01:18:41,056 --> 01:18:45,127 FOR BOTH CONFIRMATIONS OF FOLD 1687 01:18:45,127 --> 01:18:47,229 SWICHRS WOULD HAVE BEEN IN 1688 01:18:47,229 --> 01:18:48,564 TRAINING SETS. 1689 01:18:48,564 --> 01:18:50,032 WHERE DO WE GO? 1690 01:18:50,032 --> 01:18:51,567 IS IT POSSIBLE TO PREDICT FOLD 1691 01:18:51,567 --> 01:18:53,369 SWITCHING FROM SEQUENCE? 1692 01:18:53,369 --> 01:18:55,070 BEFORE ALPHAFOLD WAS DEVELOPED 1693 01:18:55,070 --> 01:18:58,273 MY LAB DEVELOPED A METHOD TO 1694 01:18:58,273 --> 01:18:58,874 PREDICT FOLD SWITCHING FOR THE 1695 01:18:58,874 --> 01:19:00,042 FIRST TIME IN A FAMILY OF 1696 01:19:00,042 --> 01:19:03,212 PROTEINS AND THIS IS AGAIN THE 1697 01:19:03,212 --> 01:19:07,182 RFAH TRANSCRIPTIONAL REGULATOR. 1698 01:19:07,182 --> 01:19:11,820 AGAIN N-TERMINAL DOMAIN IS THE 1699 01:19:11,820 --> 01:19:12,788 SAME, C SWITCHES. 1700 01:19:12,788 --> 01:19:15,691 WHAT'S INTERESTING IT HAS 1701 01:19:15,691 --> 01:19:18,527 NUMEROUS HOMOLOGS THAT ONLY 1702 01:19:18,527 --> 01:19:19,695 ASSUME ALPHA HELICAL 1703 01:19:19,695 --> 01:19:25,601 CONFIRMATION, THEY ARE SINGLE 1704 01:19:25,601 --> 01:19:27,469 FOLDING PROTEINOME BETA SHEET. 1705 01:19:27,469 --> 01:19:29,905 COULD WE DISTINGUISH BETWEEN 1706 01:19:29,905 --> 01:19:31,573 SEQUENCES THAT SWITCH, BETWEEN 1707 01:19:31,573 --> 01:19:36,045 HELIX AND BETA SHEET AND ONESOE 1708 01:19:36,045 --> 01:19:38,113 BETA SHEET. 1709 01:19:38,113 --> 01:19:41,383 I CAN'T GET INTO DETAILS USING 1710 01:19:41,383 --> 01:19:42,284 CONSERVATION PATTERNS WITH A 1711 01:19:42,284 --> 01:19:43,819 SECONDARY PREDICTOR AND WERE 1712 01:19:43,819 --> 01:19:46,855 ABLE TO BE SUCCESSFUL. 1713 01:19:46,855 --> 01:19:51,160 SO THIS IS NETWORK DIAGRAM OF 1714 01:19:51,160 --> 01:19:52,061 ALL 15,000 SEQUENCES FROM THE 1715 01:19:52,061 --> 01:19:56,098 RFA IN THE SHEET FAMILY, NODES 1716 01:19:56,098 --> 01:19:57,166 REPRESENT CLUSTERS OF SEQUENCES, 1717 01:19:57,166 --> 01:19:59,234 THE SIZE OF THE NODES HAVE TO DO 1718 01:19:59,234 --> 01:20:00,803 WITH HOW MANY SEQUENCES ARE IN 1719 01:20:00,803 --> 01:20:04,006 EACH CLUSTER, SO THE SMALLEST 1720 01:20:04,006 --> 01:20:06,942 HAVE 75% OR FEWER, LARGEST HAS A 1721 01:20:06,942 --> 01:20:08,811 LITTLE OVER 1100, CONNECTED BY 1722 01:20:08,811 --> 01:20:10,079 EDGES HAVE AN AVERAGE SEQUENCE 1723 01:20:10,079 --> 01:20:16,218 IDENTITY OF AT LEAST 24%, TAKEN 1724 01:20:16,218 --> 01:20:17,286 FROM SOMETHING. 1725 01:20:17,286 --> 01:20:20,255 WE PICKED SEQUENCES FROM MANY 1726 01:20:20,255 --> 01:20:21,623 CLUSTERS WITH MANY PREDICTIONS, 1727 01:20:21,623 --> 01:20:22,858 AVERAGE PAIR-WIDE SEQUENCE 1728 01:20:22,858 --> 01:20:24,960 IDENTITY BETWEEN THE TWO 1729 01:20:24,960 --> 01:20:26,061 SEQUENCES WAS 35% BECAUSE WE 1730 01:20:26,061 --> 01:20:28,931 WANTED TO TEST FOR ROBUSTNESS. 1731 01:20:28,931 --> 01:20:36,038 WE DID IN THE LAB CIRCULAR 1732 01:20:36,038 --> 01:20:38,207 MEASUREMENTS, YOU CAN SEE 1733 01:20:38,207 --> 01:20:39,942 PROTEINS PREDICTED TO SWITCH 1734 01:20:39,942 --> 01:20:41,844 FOLDS DIFFERENT PROFILE THAN 1735 01:20:41,844 --> 01:20:43,145 THOSE SINGLE FOLDERS AND ALSO 1736 01:20:43,145 --> 01:20:46,548 ALPHA HELIX TO BETA STRAND 1737 01:20:46,548 --> 01:20:48,784 RATIOS SHOWED SEPARATION. 1738 01:20:48,784 --> 01:20:50,419 WE DID FOLLOW-UP NMR 1739 01:20:50,419 --> 01:20:52,121 EXPERIMENTS, ASSIGNED A COUPLE 1740 01:20:52,121 --> 01:20:52,821 AND LOOKED AT SECONDARY 1741 01:20:52,821 --> 01:20:53,689 STRUCTURES, NOT SHOWING THAT ON 1742 01:20:53,689 --> 01:20:55,224 THE SLIDE BUT THOSE WERE 1743 01:20:55,224 --> 01:20:56,892 CONSISTENT WITH WHAT WE WERE 1744 01:20:56,892 --> 01:20:57,359 FINDING. 1745 01:20:57,359 --> 01:20:59,027 NOW, THE INTERESTING THING FOR 1746 01:20:59,027 --> 01:21:00,829 FUTURE PREDICTIONS AND MORE 1747 01:21:00,829 --> 01:21:01,697 GENERALIZABLE PREDICTIONS IS 1748 01:21:01,697 --> 01:21:03,532 WHEN WE LOOKED AT OUR SIX 1749 01:21:03,532 --> 01:21:05,000 PREDICTED TO SWITCH FOLDS WHICH 1750 01:21:05,000 --> 01:21:06,602 MEANS THEY WOULD HAVE A GROUND 1751 01:21:06,602 --> 01:21:08,904 STATE OF HELICAL CONFIRMATION 1752 01:21:08,904 --> 01:21:10,806 AND WE TESTED THEM AGAINST THE 1753 01:21:10,806 --> 01:21:12,341 MOST ADVANCED METHODS AT THE 1754 01:21:12,341 --> 01:21:14,276 TIME, WE FOUND THAT ALL OF THE 1755 01:21:14,276 --> 01:21:16,245 DIFFERENT METHODS THAT WERE 1756 01:21:16,245 --> 01:21:19,414 PREDICTING THIS GROUND STATE OF 1757 01:21:19,414 --> 01:21:20,849 THESE PROTEINS ONLY PREDICTED 1758 01:21:20,849 --> 01:21:23,018 BETA SHEET. 1759 01:21:23,018 --> 01:21:26,922 THE ONLY EXCEPTION WAS E. COLI 1760 01:21:26,922 --> 01:21:28,757 RFAH FOR ALPHAFOLD2, THAT 1761 01:21:28,757 --> 01:21:30,893 STRUCTURE WAS PRESUMABLY IN ITS 1762 01:21:30,893 --> 01:21:31,593 TRAINING SET. 1763 01:21:31,593 --> 01:21:35,130 SO THEN THE QUESTION IS, SO CAN 1764 01:21:35,130 --> 01:21:37,099 WE MAKE ALPHAFOLD2 WORK BETTER 1765 01:21:37,099 --> 01:21:38,500 SO IT CAN MAKE MORE ROBUST 1766 01:21:38,500 --> 01:21:40,035 PREDICTIONS OF THE SORTS OF 1767 01:21:40,035 --> 01:21:41,003 PROTEINS THAT WE'RE INTERESTED 1768 01:21:41,003 --> 01:21:41,436 IN? 1769 01:21:41,436 --> 01:21:43,872 TO BE ABLE TO ANSWER THAT 1770 01:21:43,872 --> 01:21:48,777 QUESTION WE NEED TO UNDERSTAND 1771 01:21:48,777 --> 01:21:50,345 HOW ALPHAFOLD2 WORKS. 1772 01:21:50,345 --> 01:21:51,947 IF WE HAVE INPUT SEQUENCE IN 1773 01:21:51,947 --> 01:21:54,516 GENETIC DATABASE, WE MAKE A 1774 01:21:54,516 --> 01:21:56,051 MULTIPLE SEQUENCE ALIGNMENT OR 1775 01:21:56,051 --> 01:21:59,655 MSA, WHICH I'LL TALK ABOUT IN 1776 01:21:59,655 --> 01:22:00,189 THIS TALK. 1777 01:22:00,189 --> 01:22:02,591 ALPHAFOLD WILL MAKE A 1778 01:22:02,591 --> 01:22:04,393 REPRESENTATION, ALSO A PAIR 1779 01:22:04,393 --> 01:22:05,127 REPRESENTATION, THAT'S 1780 01:22:05,127 --> 01:22:06,862 INITIALIZED AS BASICALLY NOISE, 1781 01:22:06,862 --> 01:22:10,499 THAT'S PUT INTO A BLOCK THAT 1782 01:22:10,499 --> 01:22:12,668 THEN TAKES MULTIPLE SEQUENCE 1783 01:22:12,668 --> 01:22:14,670 ALIGNMENT AND PAIR 1784 01:22:14,670 --> 01:22:16,238 REPRESENTATION AND INFERS 1785 01:22:16,238 --> 01:22:18,607 PAIRWISE DISTANCE RESTRAINTS 1786 01:22:18,607 --> 01:22:22,411 THAT ARE USED PASSED INTO THE 1787 01:22:22,411 --> 01:22:24,947 STRUCTURAL MODULE WITH A SINGLE 1788 01:22:24,947 --> 01:22:26,381 SEQUENCE REPRESENTATION TO 1789 01:22:26,381 --> 01:22:27,149 OUTPUT 3D MODEL. 1790 01:22:27,149 --> 01:22:29,785 THIS IS THE INITIAL PASS THROUGH 1791 01:22:29,785 --> 01:22:30,853 ALPHAFOLD NETWORK, PASSED BACK 1792 01:22:30,853 --> 01:22:34,189 ALONG WITH THE UPDATED PAIR 1793 01:22:34,189 --> 01:22:35,023 REPRESENTATION, AND RECYCLED 1794 01:22:35,023 --> 01:22:36,325 NUMEROUS TIMES TO IMPROVE 1795 01:22:36,325 --> 01:22:37,092 INITIAL PREDICTION. 1796 01:22:37,092 --> 01:22:39,228 WHAT I WANT TO EMPHASIZE HERE IS 1797 01:22:39,228 --> 01:22:43,265 THAT EVEN THOUGH THIS IS KIND OF 1798 01:22:43,265 --> 01:22:45,634 AN EXPLANATION AT HIGH LEVEL HOW 1799 01:22:45,634 --> 01:22:46,635 ALPHAFOLD IS WORKING, IT'S STILL 1800 01:22:46,635 --> 01:22:48,103 BECAUSE IT'S A DEEP LEARNING 1801 01:22:48,103 --> 01:22:49,805 NETWORK A BLACK BOX. 1802 01:22:49,805 --> 01:22:50,806 WE DON'T REALLY KNOW WHAT'S 1803 01:22:50,806 --> 01:22:52,875 HAPPENING UNDER THE HOOD AND 1804 01:22:52,875 --> 01:22:54,810 WHAT ALPHAFOLD IS SEEING THAT'S 1805 01:22:54,810 --> 01:22:56,245 ALLOWING IS TO MAKE INFERENCES. 1806 01:22:56,245 --> 01:22:57,446 THIS IS AN IMPORTANT QUESTION 1807 01:22:57,446 --> 01:22:59,448 FOR MY LAB BECAUSE IF WE'RE 1808 01:22:59,448 --> 01:23:01,416 GOING TO BUILD A MODEL OR 1809 01:23:01,416 --> 01:23:02,484 IMPROVE UPON AN EXISTING MODEL 1810 01:23:02,484 --> 01:23:04,720 TO REALLY BE ABLE TO PREDICT 1811 01:23:04,720 --> 01:23:05,687 FOLD SWITCHING ROBUSTLY WE NEED 1812 01:23:05,687 --> 01:23:07,256 TO KNOW HOW IT WORKS. 1813 01:23:07,256 --> 01:23:11,393 NOW, IF YOU READ THE NOBEL PRIZE 1814 01:23:11,393 --> 01:23:14,496 ANNOUNCEMENT FROM 2024, THE 1815 01:23:14,496 --> 01:23:16,431 CLAIM IS THAT CO-EVOLUTION IS A 1816 01:23:16,431 --> 01:23:19,501 REALLY IMPORTANT PART IN DRIVING 1817 01:23:19,501 --> 01:23:21,236 ALPHAFOLD BASE PREDICTIONS, 1818 01:23:21,236 --> 01:23:24,072 BASICALLY WHEN YOU HAVE PAIRS OF 1819 01:23:24,072 --> 01:23:26,742 AMINO ACIDS THAT VARY AT A 1820 01:23:26,742 --> 01:23:30,646 STATISTICALLY SIGNIFICANT LEVEL 1821 01:23:30,646 --> 01:23:31,780 AND WHEN THAT HAPPENS THOSE 1822 01:23:31,780 --> 01:23:33,815 AMINO ACIDS HAVE BEEN FOUND TO 1823 01:23:33,815 --> 01:23:35,717 BE OFTEN VERY CLOSE IN SPACE. 1824 01:23:35,717 --> 01:23:38,987 SO THESE CAN BE -- CO-EVOLVING 1825 01:23:38,987 --> 01:23:40,989 RESIDUES CAN BE USED TO INFORM 1826 01:23:40,989 --> 01:23:42,925 DISTANCE MAP OR THE RESTRAINTS 1827 01:23:42,925 --> 01:23:45,394 THAT ALPHAFOLD IS USING TO BE 1828 01:23:45,394 --> 01:23:47,496 ABLE TO PREDICT THREE 1829 01:23:47,496 --> 01:23:48,764 DIMENSIONAL STRUCTURED 1830 01:23:48,764 --> 01:23:49,064 APPROACHES. 1831 01:23:49,064 --> 01:23:50,732 INDEED I'M SURE THIS IS HOW 1832 01:23:50,732 --> 01:23:52,200 ALPHAFOLD WORKS MUCH OF THE 1833 01:23:52,200 --> 01:23:52,401 TIME. 1834 01:23:52,401 --> 01:23:54,102 BUT WHAT I'M GOING TO TELL YOU 1835 01:23:54,102 --> 01:23:56,705 NOW IS THAT IS NOT HOW IT WORKS 1836 01:23:56,705 --> 01:24:03,745 FOR FOLD SWITCHING PROTEINS. 1837 01:24:03,745 --> 01:24:05,981 WHY IS CO-EVOLUTION NOT DRIVING 1838 01:24:05,981 --> 01:24:06,682 PREDICTIONS OF ALTERNATIVE 1839 01:24:06,682 --> 01:24:08,150 PROTEINS OF FOLD SWITCHING? 1840 01:24:08,150 --> 01:24:11,486 THIS COMES FROM STUDIES WE HAVE 1841 01:24:11,486 --> 01:24:14,256 DONE PREVIOUSLY WHERE WE LOOKED 1842 01:24:14,256 --> 01:24:17,993 AT ALPHAFOLD PREDICTIONS OF 1843 01:24:17,993 --> 01:24:23,065 FAVORITE PROTEIN RFAH, THIS 1844 01:24:23,065 --> 01:24:24,700 BLACK PAIRWISE DISTANCES ARE 1845 01:24:24,700 --> 01:24:30,872 THINGS WITHIN ANGSTROMS OF ONE 1846 01:24:30,872 --> 01:24:31,873 ANOTHER. 1847 01:24:31,873 --> 01:24:34,042 IF WE COMPARE TO CONTACT UNIQUE 1848 01:24:34,042 --> 01:24:36,812 TO THE BETA SHEET FOLD ON THE 1849 01:24:36,812 --> 01:24:38,780 LOWER DIAGONAL, THOSE ARE BLUE, 1850 01:24:38,780 --> 01:24:41,516 AND UNIQUE TO ALPHA HELICAL 1851 01:24:41,516 --> 01:24:42,584 CONFIRMATION WHICH UPPER 1852 01:24:42,584 --> 01:24:46,888 DIAGONAL ARE YELLOW, YOU CAN SEE 1853 01:24:46,888 --> 01:24:48,991 THERE'S GOOD CORRESPONDENCE 1854 01:24:48,991 --> 01:24:52,027 BETWEEN WHAT IS SEEN, 1855 01:24:52,027 --> 01:24:53,462 CO-EVOLUTIONARY INFORMATION, AND 1856 01:24:53,462 --> 01:24:54,930 BETA SHEET CONFIRMATION, NOT 1857 01:24:54,930 --> 01:24:55,964 GOOD CORRESPONDENCE BETWEEN WHAT 1858 01:24:55,964 --> 01:24:58,834 IT'S SEEING WITH THE ALPHA 1859 01:24:58,834 --> 01:25:00,802 HELICAL CONFIRMATION. 1860 01:25:00,802 --> 01:25:02,838 THAT'S TRUE FOR THIS MULTIPLE 1861 01:25:02,838 --> 01:25:04,373 SEQUENCE ALIGNMENT WHICH WE 1862 01:25:04,373 --> 01:25:05,407 EXTRACTED FROM THE ALPHAFOLD 1863 01:25:05,407 --> 01:25:07,142 MODEL SO WE KNOW THAT'S WHAT IT 1864 01:25:07,142 --> 01:25:11,079 USED TO MAKE THESE PREDICTIONS, 1865 01:25:11,079 --> 01:25:14,483 BUT IT IS TRUE, SORRY, IT IS 1866 01:25:14,483 --> 01:25:16,284 TRUE FOR ALL THE MULTIPLE 1867 01:25:16,284 --> 01:25:17,152 SEQUENCE ALIGNMENTS THAT IT'S 1868 01:25:17,152 --> 01:25:19,921 USING TO BE ABLE TO MAKE ITS 1869 01:25:19,921 --> 01:25:20,222 PREDICTION. 1870 01:25:20,222 --> 01:25:22,991 THEY ALL MATCH UP BETTER WITH OF 1871 01:25:22,991 --> 01:25:24,826 THE BETA SHEET THAN WITH THE 1872 01:25:24,826 --> 01:25:26,928 ALPHA HELIX WHEN WE ONLY DO ONE 1873 01:25:26,928 --> 01:25:29,664 PASS THROUGH THE MODEL. 1874 01:25:29,664 --> 01:25:32,601 INTERESTINGLY, IT'S THE 1875 01:25:32,601 --> 01:25:34,136 RECYCLING PROCESS THAT LEADS TO 1876 01:25:34,136 --> 01:25:36,204 THE CHANGE, SO INITIALLY IF YOU 1877 01:25:36,204 --> 01:25:38,306 PASS AN MSA INTO ALPHAFOLD IT 1878 01:25:38,306 --> 01:25:41,810 WILL PREDICT THIS BETA SHEET BUT 1879 01:25:41,810 --> 01:25:44,880 AS YOU RECYCLE WITH THE EXACTLY 1880 01:25:44,880 --> 01:25:46,515 NSA WE SHOWED WITH THE SAME 1881 01:25:46,515 --> 01:25:48,683 INFORMATION FOR BETA SHEET, 1882 01:25:48,683 --> 01:25:50,552 EVENTUALLY BECOMES HELICAL. 1883 01:25:50,552 --> 01:25:52,087 WE THOUGHT THAT'S STRANGE, HOW 1884 01:25:52,087 --> 01:25:53,555 IS IT INFERRING HELIX FROM 1885 01:25:53,555 --> 01:25:56,091 SOMETHING THAT HAS A VERY STRONG 1886 01:25:56,091 --> 01:25:56,825 BETA SHEET CONFIRMATION? 1887 01:25:56,825 --> 01:25:59,861 AND SO WE WERE ABLE TO -- OUR 1888 01:25:59,861 --> 01:26:01,897 THOUGHT WAS THIS HAS MORE TO DO 1889 01:26:01,897 --> 01:26:03,632 WITH LEARNING OF PROTEIN 1890 01:26:03,632 --> 01:26:04,699 STRUCTURE AND CO-EVOLUTIONARY 1891 01:26:04,699 --> 01:26:07,202 INFERENCE. 1892 01:26:07,202 --> 01:26:09,938 WE TESTED THIS DIRECTLY BECAUSE 1893 01:26:09,938 --> 01:26:11,773 ANOTHER ALPHAFOLD MODEL, MODEL 1894 01:26:11,773 --> 01:26:15,243 BASED ON ARCHITECTURE USING SAME 1895 01:26:15,243 --> 01:26:18,380 PROTOCOL, WAS DEVELOPED CALLED 1896 01:26:18,380 --> 01:26:18,780 C-FOLD. 1897 01:26:18,780 --> 01:26:20,315 THE ONLY DIFFERENCE IS THE WAY 1898 01:26:20,315 --> 01:26:21,116 IT WAS TRAINED. 1899 01:26:21,116 --> 01:26:23,552 SO IT WAS TRAINED ON A 1900 01:26:23,552 --> 01:26:26,788 CONFIRMATIONAL SPLIT OF THE PDP 1901 01:26:26,788 --> 01:26:28,957 SO NOT ALL ALTERNATIVE 1902 01:26:28,957 --> 01:26:30,025 CONFIRMATIONS WERE THERE, THE 1903 01:26:30,025 --> 01:26:32,727 DEVELOPER PATRICK BRYANT SENT ME 1904 01:26:32,727 --> 01:26:34,529 THE LIST, NO RFA HELICAL 1905 01:26:34,529 --> 01:26:35,797 STRUCTURES IN THERE. 1906 01:26:35,797 --> 01:26:38,867 THEN WE CAN PASS IN THE SAME 1907 01:26:38,867 --> 01:26:40,402 MULTIPLE SEQUENCE ALIGNMENT 1908 01:26:40,402 --> 01:26:43,472 THAT'S USED BY ALPHAFOLD TO GET 1909 01:26:43,472 --> 01:26:45,040 HELICAL CON FIRMATION, PASSION 1910 01:26:45,040 --> 01:26:47,042 INTO C-FOLD, END UP GETTING THE 1911 01:26:47,042 --> 01:26:48,410 BETA SHEET CONFIRMATION. 1912 01:26:48,410 --> 01:26:52,147 THIS STRONGLY SUGGESTS THIS 1913 01:26:52,147 --> 01:26:55,016 HELICAL CONFIRMATION IS INFORMED 1914 01:26:55,016 --> 01:26:57,986 BY ALPHAFOLD TRAINING RATHER 1915 01:26:57,986 --> 01:27:01,890 THAN CO-EVOLUTIONARY INFERENCE. 1916 01:27:01,890 --> 01:27:05,727 THIS MAY EXPLAIN WHY IT 1917 01:27:05,727 --> 01:27:08,230 STRUGGLED TO PREDICT NEW PROTEIN 1918 01:27:08,230 --> 01:27:08,597 FOLDS CORRECTLY. 1919 01:27:08,597 --> 01:27:11,666 WHEN YOU PASS THIS INTO 1920 01:27:11,666 --> 01:27:14,769 ALPHAFOLD, IT PREDICTS THE 1921 01:27:14,769 --> 01:27:23,044 STRUCTURE OF BCCIP BETA, WITHIN 1922 01:27:23,044 --> 01:27:24,579 2 ANGSTROMS, BUT 13 DIFFERENT 1923 01:27:24,579 --> 01:27:28,717 FROM THE CORRECT STRUCTURE. 1924 01:27:28,717 --> 01:27:30,919 HUMAN CO-INTERLEUKIN-2, AN 1925 01:27:30,919 --> 01:27:33,455 IMPORTANT PROTEIN, SOLVED BY 1926 01:27:33,455 --> 01:27:37,092 LOUIS CAGE THIS SUMMER. 1927 01:27:37,092 --> 01:27:37,959 THREE DIMENSIONAL ALPHAFOLD 1928 01:27:37,959 --> 01:27:40,228 PREDICTS MATCHES SOMETHING IN 1929 01:27:40,228 --> 01:27:44,299 THE TRAINING SAID BUT DOES NOT 1930 01:27:44,299 --> 01:27:46,801 MATCH WITH ANGSTROMS OFF. 1931 01:27:46,801 --> 01:27:49,504 FURTHERMORE, A DNA BINDING 1932 01:27:49,504 --> 01:27:53,575 PROTEIN AN SAYS ANCESTOR MATCHE 1933 01:27:53,575 --> 01:27:56,378 TRAINING SET BUT NOT 1934 01:27:56,378 --> 01:27:56,678 EXPERIMENTS. 1935 01:27:56,678 --> 01:27:59,047 CAN WE LEVERAGE THIS KNOWLEDGE 1936 01:27:59,047 --> 01:28:02,984 THAT WE HAVE, IF ALPHAFOLD IS 1937 01:28:02,984 --> 01:28:05,053 FOR ALTERNATIVE CONFIRMATIONS 1938 01:28:05,053 --> 01:28:06,821 MAKING ASSOCIATIONS IN ITS TRAIN 1939 01:28:06,821 --> 01:28:09,424 SET CAN WE LEVERAGE TO DO THAT 1940 01:28:09,424 --> 01:28:09,991 BETTER. 1941 01:28:09,991 --> 01:28:13,295 WE DEVELOPED A METHOD CALLED CF 1942 01:28:13,295 --> 01:28:14,996 RANDOM, WE HAVE INPUT SEQUENCE 1943 01:28:14,996 --> 01:28:16,097 AND FULL LENGTH MULTIPLE 1944 01:28:16,097 --> 01:28:17,332 SEQUENCE ALIGNMENT. 1945 01:28:17,332 --> 01:28:19,401 WE FROM THIS EXPECT WE'LL GET 1946 01:28:19,401 --> 01:28:20,368 THE DOMINANT CONFIRMATION 1947 01:28:20,368 --> 01:28:20,936 WITHOUT A PROBLEM. 1948 01:28:20,936 --> 01:28:23,138 THEN THE QUESTION IS HOW ARE WE 1949 01:28:23,138 --> 01:28:24,439 GOING TO GET ALTERNATIVE 1950 01:28:24,439 --> 01:28:24,906 CONFIRMATION. 1951 01:28:24,906 --> 01:28:28,143 TO BE ABLE TO GET THAT WE 1952 01:28:28,143 --> 01:28:30,345 GENERATE EXTREMELY SHALLOW 1953 01:28:30,345 --> 01:28:32,747 MULTIPLE SEQUENCE ALIGNMENTS TOO 1954 01:28:32,747 --> 01:28:34,049 SHALLOW FOR CO-EVOLUTIONARY 1955 01:28:34,049 --> 01:28:35,250 INFERENCE, THEY LIKE THE 1956 01:28:35,250 --> 01:28:35,550 RESEQUENCES. 1957 01:28:35,550 --> 01:28:38,453 WHAT WE THINK WE'RE DOING IS 1958 01:28:38,453 --> 01:28:40,188 JOGGING ALPHAFOLD MEMORY TO MAKE 1959 01:28:40,188 --> 01:28:41,189 ASSOCIATIONS WITH WHAT'S LEARNED 1960 01:28:41,189 --> 01:28:42,357 DURING TRAINING SO WE CAN RUN IT 1961 01:28:42,357 --> 01:28:44,893 ON A BUNCH OF SHALLOW RANDOM 1962 01:28:44,893 --> 01:28:46,928 SEQUENCE ALIGNMENTS AND IN AN 1963 01:28:46,928 --> 01:28:50,565 UNBIASED WAY SELECT THE MOST 1964 01:28:50,565 --> 01:28:53,134 LIKELY PROBABLE ALTERNATIVE 1965 01:28:53,134 --> 01:28:53,468 CONFIRMATION. 1966 01:28:53,468 --> 01:28:55,604 NOW WE CAN COMPARE CF RANDOM TO 1967 01:28:55,604 --> 01:28:58,473 ANOTHER METHOD FOR PREDICTING 1968 01:28:58,473 --> 01:28:59,241 ALTERNATIVE CONFIRMATIONS CALLED 1969 01:28:59,241 --> 01:29:01,543 AF CLUSTER WHICH CLAIMS TO WORK 1970 01:29:01,543 --> 01:29:02,143 BY CO-EVOLUTIONARY INFERENCE, 1971 01:29:02,143 --> 01:29:04,346 AND THE WAY THEY SAY IT WORKS IS 1972 01:29:04,346 --> 01:29:06,915 THAT YOU HAVE THESE SEQUENCED 1973 01:29:06,915 --> 01:29:10,719 CLUSTERS THAT CONTAIN ROBUST 1974 01:29:10,719 --> 01:29:12,721 CO-EVOLUTIONARY INFORMATION THAT 1975 01:29:12,721 --> 01:29:13,321 ENCODE ALTERNATIVE PROTEIN 1976 01:29:13,321 --> 01:29:13,855 CO-FOLDS. 1977 01:29:13,855 --> 01:29:16,191 YOU EXPECT IF WE RUN ALPHAFOLD 1978 01:29:16,191 --> 01:29:17,859 AND GIVING IT CONCERTED 1979 01:29:17,859 --> 01:29:18,827 INFORMATION FOR THE ALTERNATIVE 1980 01:29:18,827 --> 01:29:20,495 CONFIRMATION WHICH IS WHAT AF 1981 01:29:20,495 --> 01:29:22,631 CLUSTER CLAIMS THAT WOULD DO 1982 01:29:22,631 --> 01:29:26,334 BETTER THAN GIVING IT NOISY 1983 01:29:26,334 --> 01:29:28,236 RANDOM SHALLOW INFORMATION. 1984 01:29:28,236 --> 01:29:32,274 BUT ACTUALLY THE OPPOSITE IS 1985 01:29:32,274 --> 01:29:32,507 TRUE. 1986 01:29:32,507 --> 01:29:33,908 CF RANDOM OUTPERFORMS. 1987 01:29:33,908 --> 01:29:36,544 ALL THE EXAMPLES OF THE 1988 01:29:36,544 --> 01:29:37,746 STRUCTURES PREDICTED BY 1989 01:29:37,746 --> 01:29:39,814 AF-CLUSTER AND HAVE BEEN 1990 01:29:39,814 --> 01:29:40,782 EXPERIMENTALLY CONFIRMED, SO IN 1991 01:29:40,782 --> 01:29:43,118 THESE CASES YOU CAN SEE THAT 1992 01:29:43,118 --> 01:29:46,421 PRETTY MUCH ALLS CF RANDOM IS 1993 01:29:46,421 --> 01:29:48,056 MORE -- PRODUCES MORE ACCURATE 1994 01:29:48,056 --> 01:29:50,025 PREDICTS THAN AF CLUSTER OFTEN 1995 01:29:50,025 --> 01:29:52,227 WITH HIGHER CONFIDENCE, 1996 01:29:52,227 --> 01:29:53,662 ESPECIALLY FOR ALTERNATIVE 1997 01:29:53,662 --> 01:29:54,362 CONFIRMATIONS. 1998 01:29:54,362 --> 01:29:56,364 NOT ONLY IS AF CLUSTER MORE 1999 01:29:56,364 --> 01:29:58,366 ACCURATE BUT OVERALL WHEN WE 2000 01:29:58,366 --> 01:30:02,837 LOOK AT THE ENSEMBLE WE SEE IT 2001 01:30:02,837 --> 01:30:04,072 PRODUCES A BETTER COEFFICIENT IN 2002 01:30:04,072 --> 01:30:07,609 ALL CASES OF THE PROTEIN 2003 01:30:07,609 --> 01:30:08,610 FAMILIES TESTED, MUCH HIGHER 2004 01:30:08,610 --> 01:30:11,246 SUCCESS RATE, NOT ONLY MORE 2005 01:30:11,246 --> 01:30:14,215 ACCURATE AND SUCCESSFUL BUT ALSO 2006 01:30:14,215 --> 01:30:15,216 MUCH MORE EFFICIENT. 2007 01:30:15,216 --> 01:30:20,955 WE ONLY NEED TO DO A FEW RUNS ON 2008 01:30:20,955 --> 01:30:23,158 CHI MAG 2 AND IFH COMPARED TO 2009 01:30:23,158 --> 01:30:25,260 HUNDREDS FOR AF CLUSTER AND 2010 01:30:25,260 --> 01:30:27,529 NUMBER OF PREDICTIONS NEEDED IS 2011 01:30:27,529 --> 01:30:29,331 SUBSTANTIALLY FEWER TO REALIZE 2012 01:30:29,331 --> 01:30:30,131 ALTERNATIVE CONFIRMATION. 2013 01:30:30,131 --> 01:30:31,933 NOW WE COME BACK TO THE QUESTION 2014 01:30:31,933 --> 01:30:34,836 OF CO-EVOLUTION AGAIN AND ASK IS 2015 01:30:34,836 --> 01:30:37,605 THERE SOMETHING WE MISSING, IS 2016 01:30:37,605 --> 01:30:38,807 THERE CO-EVOLUTIONARY 2017 01:30:38,807 --> 01:30:40,775 INFORMATION USED BY CF RANDOM? 2018 01:30:40,775 --> 01:30:42,844 WE EXTRACTED THE EXACT MSAs 2019 01:30:42,844 --> 01:30:45,380 THAT CF RANDOM WAS USING TO 2020 01:30:45,380 --> 01:30:47,248 PREDICT, AND HERE IS WHAT WE 2021 01:30:47,248 --> 01:30:48,917 GOT. 2022 01:30:48,917 --> 01:30:51,553 IN PINK THIS IS THE DEPTH WE 2023 01:30:51,553 --> 01:30:53,588 SAMPLED THAT GAVE US ALPHA 2024 01:30:53,588 --> 01:30:54,356 HELICAL CONFIRMATION, THAT'S 2025 01:30:54,356 --> 01:30:55,990 LIKE IN A PINK BOX. 2026 01:30:55,990 --> 01:30:59,227 AND YOU CAN SEE WE'VE GOT OUR 2027 01:30:59,227 --> 01:31:01,930 COMMON CONTACTS, THE ONES UNIQUE 2028 01:31:01,930 --> 01:31:04,199 TO THE SHARED N-TERMINAL DOMAIN. 2029 01:31:04,199 --> 01:31:06,534 WE'VE GOT DOMINANT CONTACTS, 2030 01:31:06,534 --> 01:31:07,836 UNIQUE TO BETA SHEET, 2031 01:31:07,836 --> 01:31:09,137 ALTERNATIVE CONTACTS UNIQUE TO 2032 01:31:09,137 --> 01:31:09,437 ALPHA HELIX. 2033 01:31:09,437 --> 01:31:10,538 WHAT I WANT TO DRAW YOUR 2034 01:31:10,538 --> 01:31:13,608 ATTENTION TO HERE IS THAT EVEN 2035 01:31:13,608 --> 01:31:15,243 THOUGH THE ALPHA HELICAL 2036 01:31:15,243 --> 01:31:18,213 CONFIRMATION IS BEING PREDICTED, 2037 01:31:18,213 --> 01:31:19,247 THERE'S ESSENTIALLY NO 2038 01:31:19,247 --> 01:31:19,948 CO-EVOLUTIONARY INFORMATION 2039 01:31:19,948 --> 01:31:22,450 UNIQUE TO IT. 2040 01:31:22,450 --> 01:31:23,451 AT ANY SAMPLING. 2041 01:31:23,451 --> 01:31:25,553 THIS AGAIN SUGGESTS THAT 2042 01:31:25,553 --> 01:31:28,256 CO-EVOLUTION IS NOT DRIVING THIS 2043 01:31:28,256 --> 01:31:28,957 PREDICTION OF ALTERNATIVE 2044 01:31:28,957 --> 01:31:31,559 CONFIRMATION, TRUE NOT ONLY FOR 2045 01:31:31,559 --> 01:31:35,730 RFAH BUT ALSO TRUE FOR MAG 2 AND 2046 01:31:35,730 --> 01:31:36,698 CHI B. 2047 01:31:36,698 --> 01:31:38,933 THERE'S NO PINK INFORMATION. 2048 01:31:38,933 --> 01:31:40,235 THERE'S ACTUALLY MORE BLUE 2049 01:31:40,235 --> 01:31:42,404 INFORMATION THAN PINK. 2050 01:31:42,404 --> 01:31:43,505 SO, THAT'S INTERESTED THAT THIS 2051 01:31:43,505 --> 01:31:45,473 IS THE WAY IT SEEMS TO BE 2052 01:31:45,473 --> 01:31:47,208 WORKING BUT I'VE ONLY SHOWN 2053 01:31:47,208 --> 01:31:49,077 THREE EXAMPLES. 2054 01:31:49,077 --> 01:31:51,513 NEXT NATURAL QUESTION IS HOW 2055 01:31:51,513 --> 01:31:53,648 WELL DOES CF RANDOM GENERALIZE 2056 01:31:53,648 --> 01:31:55,250 ON FOLD SWICHRS? 2057 01:31:55,250 --> 01:31:57,886 AND WE DID THAT TEST, YOU CAN 2058 01:31:57,886 --> 01:32:01,489 SEE THAT CF RANDOM NOT ONLY 2059 01:32:01,489 --> 01:32:03,224 PREDICTS MORE SUCCESSFUL 2060 01:32:03,224 --> 01:32:08,496 PREDICTIONS, IT REQUIRES 2061 01:32:08,496 --> 01:32:09,697 SUBSTANTIALLY LESS SAMPLING, ALL 2062 01:32:09,697 --> 01:32:11,466 THE METHODS WITHOUT TEMPLATES, 2063 01:32:11,466 --> 01:32:12,867 WE NEEDED TO SAMPLE 30,000 2064 01:32:12,867 --> 01:32:15,837 STRUCTURES AS OPPOSED TO OVER 2065 01:32:15,837 --> 01:32:16,704 200,000. 2066 01:32:16,704 --> 01:32:20,408 AND IT TURNS OUT THAT CF RANDOM 2067 01:32:20,408 --> 01:32:22,210 GENERALIZES BEYOND FOLD 2068 01:32:22,210 --> 01:32:22,477 SWITCHING. 2069 01:32:22,477 --> 01:32:24,412 THIS IS USING THE HARDEST 2070 01:32:24,412 --> 01:32:26,080 EXAMPLE, WE'RE ABLE TO DO OTHERS 2071 01:32:26,080 --> 01:32:28,550 ALSO IMPORTANT BUT MAYBE NOT AS 2072 01:32:28,550 --> 01:32:29,284 TECHNICALLY CHALLENGING. 2073 01:32:29,284 --> 01:32:31,486 WE WERE ABLE TO BENCHMARK CF 2074 01:32:31,486 --> 01:32:32,887 RANDOM AGAINST TWO METHODS, 2075 01:32:32,887 --> 01:32:34,556 SHOWING YOU ONE HERE, AGAINST AF 2076 01:32:34,556 --> 01:32:38,726 SAMPLE 2 WHERE WE'RE USING OPEN 2077 01:32:38,726 --> 01:32:41,830 AND CLOSED CONFIRMATIONS OF 2078 01:32:41,830 --> 01:32:46,201 TRANSPORTERS OR PARAPLASMIC 2079 01:32:46,201 --> 01:32:47,168 BINDING PROTEINS. 2080 01:32:47,168 --> 01:32:49,037 THIS IS MORE ACCURATE IN 2081 01:32:49,037 --> 01:32:52,740 PREDICTING BOTH OPEN AND CLOSED 2082 01:32:52,740 --> 01:32:54,976 CONFIRMATIONS IN AF SAMPLE 2, 2083 01:32:54,976 --> 01:32:56,778 AND IT IS MUCH MORE EFFICIENT 2084 01:32:56,778 --> 01:32:59,881 MOST OF THE TIME. 2085 01:32:59,881 --> 01:33:02,083 AF SAMPLE 2 REQUIRES A THOUSAND 2086 01:33:02,083 --> 01:33:04,319 PREDICTIONS PER MODEL, WE 2087 01:33:04,319 --> 01:33:06,688 USUALLY ONLY NEED 200 TO REALIZE 2088 01:33:06,688 --> 01:33:08,389 MORE ACCURATE STRUCTURES. 2089 01:33:08,389 --> 01:33:10,925 THIS REALLY ATTESTS TO THE POWER 2090 01:33:10,925 --> 01:33:13,728 OF CF RANDOM AND TO APPROACH 2091 01:33:13,728 --> 01:33:16,030 WE'RE USING RIGHT NOW. 2092 01:33:16,030 --> 01:33:18,833 SO IF ALPHAFOLD ISN'T PREDICTING 2093 01:33:18,833 --> 01:33:22,237 THESE ALTERNATIVE CONFIRMATIONS 2094 01:33:22,237 --> 01:33:25,840 BY CO- EVOLUTIONARY INFERENCE 2095 01:33:25,840 --> 01:33:26,975 HOW IS IT WORKING? 2096 01:33:26,975 --> 01:33:29,544 WE LOOKED AT FOLD SWICHRS NOT 2097 01:33:29,544 --> 01:33:31,412 PREDICTED WELL. 2098 01:33:31,412 --> 01:33:35,149 THIS IS A PROTEIN, AN ALPHA BETA 2099 01:33:35,149 --> 01:33:35,483 CONFIRMATION. 2100 01:33:35,483 --> 01:33:38,286 WHEN WE RUN AF CLUSTER WE GET 2101 01:33:38,286 --> 01:33:39,721 THIS CONFIRMATION BACK MANY 2102 01:33:39,721 --> 01:33:43,391 TIMES, DON'T GET THREE HELICALL 2103 01:33:43,391 --> 01:33:44,592 CONFIRMATION AT ALL. 2104 01:33:44,592 --> 01:33:48,329 WHAT'S INTERESTING IF WE DO PDB 2105 01:33:48,329 --> 01:33:50,899 SEARCH OF THE SEQUENCE 2106 01:33:50,899 --> 01:33:54,369 REPRESENTED IN BLACK WE FIND 2107 01:33:54,369 --> 01:34:01,643 THREE HOMOLOGS THAT ASSUME THREE 2108 01:34:01,643 --> 01:34:04,012 ALPHA HELICAL BUNDLE STRUCTURE. 2109 01:34:04,012 --> 01:34:05,813 SO THE TARGET SEQUENCE, ONE 2110 01:34:05,813 --> 01:34:07,515 THREE ALPHA HELICAL BUNDLE 2111 01:34:07,515 --> 01:34:10,585 SEQUENCE, PASS THOSE INTO 2112 01:34:10,585 --> 01:34:15,189 ALPHAFOLD WE GET THREE ALPHA 2113 01:34:15,189 --> 01:34:16,491 HELIX BUNDLE STRUCTURES EVERY 2114 01:34:16,491 --> 01:34:17,058 TIME. 2115 01:34:17,058 --> 01:34:21,763 THIS IS WITH HIGH CONFIDENCE 2116 01:34:21,763 --> 01:34:23,431 BASED ON ALPHAFOLD'S SCORE. 2117 01:34:23,431 --> 01:34:25,266 SO THIS SUGGESTS ALPHAFOLD MAY 2118 01:34:25,266 --> 01:34:26,901 BE USING SOME SORT OF SEQUENCE 2119 01:34:26,901 --> 01:34:29,804 ASSOCIATION TO BE ABLE TO 2120 01:34:29,804 --> 01:34:30,972 PREDICT THESE ALTERNATIVE CON 2121 01:34:30,972 --> 01:34:33,741 FIRMATIONINGS OF FOLD SWITCHING 2122 01:34:33,741 --> 01:34:34,108 PROTEINS. 2123 01:34:34,108 --> 01:34:36,210 WE THOUGHT THIS WAS AN 2124 01:34:36,210 --> 01:34:37,078 INTERESTING OBSERVATION, COULD 2125 01:34:37,078 --> 01:34:38,947 WE TAKE IT BEYOND FOLD SWICHRS 2126 01:34:38,947 --> 01:34:41,816 AND USE IT FOR OTHER PROTEIN 2127 01:34:41,816 --> 01:34:43,551 FOLDS AS WELL? 2128 01:34:43,551 --> 01:34:46,254 MADELINE IN MY LAB HAS BEEN 2129 01:34:46,254 --> 01:34:47,355 HELPING TO INVESTIGATE THIS 2130 01:34:47,355 --> 01:34:49,791 QUESTION, SO WE HAVE AN UNBIASED 2131 01:34:49,791 --> 01:34:51,626 SET OF PROTEIN STRUCTURES WITH 2132 01:34:51,626 --> 01:34:59,167 50 TO 3 300 AMINO ACIDS, HAVE N 2133 01:34:59,167 --> 01:35:00,501 ABLE TO FIND SEQUENCES ACROSS 2134 01:35:00,501 --> 01:35:04,439 FAMILIES AND SIZES THAT ALLOW 2135 01:35:04,439 --> 01:35:05,306 ALPHAFOLD TO PREDICT 2136 01:35:05,306 --> 01:35:08,710 CONSISTENTLY JUST ONE PROTEIN 2137 01:35:08,710 --> 01:35:08,910 FOLD. 2138 01:35:08,910 --> 01:35:10,878 AGAIN, THIS ISN'T FOLD SWITCHING 2139 01:35:10,878 --> 01:35:11,112 ANYMORE. 2140 01:35:11,112 --> 01:35:14,282 WHAT'S GREAT IS IT ALLOWS US TO 2141 01:35:14,282 --> 01:35:15,483 BETTER UNDERSTAND WHAT'S 2142 01:35:15,483 --> 01:35:17,218 HAPPENING WITH THE MODEL. 2143 01:35:17,218 --> 01:35:27,295 SO HOW CAN WE INVESTIGATE HOW 2144 01:35:27,295 --> 01:35:28,429 ALPHAFOLD IS PREDICTING? 2145 01:35:28,429 --> 01:35:33,134 WE CAN DO A BUNCH OF RANDOM 2146 01:35:33,134 --> 01:35:33,701 ALANINE SUBSTITUTIONS, MANY 2147 01:35:33,701 --> 01:35:35,336 TIMES, AND THEN WE CAN DO 2148 01:35:35,336 --> 01:35:37,405 STRUCTURE PREDICTIONS ON ALL OF 2149 01:35:37,405 --> 01:35:38,072 THEM. 2150 01:35:38,072 --> 01:35:39,774 SO WE'RE DOING 3,000 ALANINE 2151 01:35:39,774 --> 01:35:41,209 KNOCKOUTS FOR PROTEIN AND 2152 01:35:41,209 --> 01:35:43,711 SCORING OUR STRUCTURES BY 2153 01:35:43,711 --> 01:35:43,978 ACCURACY. 2154 01:35:43,978 --> 01:35:46,147 THEN WE CAN USE A LINEAR 2155 01:35:46,147 --> 01:35:50,485 CLASSIFIER AND SELECT THE 2156 01:35:50,485 --> 01:35:51,619 HIGHEST WEIGHTS TO IDENTIFY 2157 01:35:51,619 --> 01:35:55,990 WHICH AMINO ACIDS MAY BE THE 2158 01:35:55,990 --> 01:35:57,625 MOST IMPORTANT FOR PREDICTING 2159 01:35:57,625 --> 01:35:58,826 THIS CONFIRMATION. 2160 01:35:58,826 --> 01:36:00,995 WHAT'S REALLY INTERESTING ABOUT 2161 01:36:00,995 --> 01:36:02,664 THIS APPROACH IS 70% OF THE 2162 01:36:02,664 --> 01:36:05,366 SECOND SEQUENCE TURNS OUT TO BE 2163 01:36:05,366 --> 01:36:06,501 ALANINE. 2164 01:36:06,501 --> 01:36:08,469 ONLY 30% OF THE INFORMATION IS 2165 01:36:08,469 --> 01:36:11,406 WHAT SEEMS TO BE REQUIRED TO BE 2166 01:36:11,406 --> 01:36:12,440 ABLE TO PREDICT THIS 2167 01:36:12,440 --> 01:36:14,042 CONFIRMATION THAT WE'RE 2168 01:36:14,042 --> 01:36:14,876 INTERESTED IN. 2169 01:36:14,876 --> 01:36:16,144 NOW, HERE IS WHERE THE RUBBER 2170 01:36:16,144 --> 01:36:17,111 MEETS THE ROAD. 2171 01:36:17,111 --> 01:36:19,180 WHAT HAPPENS IF WE COMPARE THE 2172 01:36:19,180 --> 01:36:21,616 FULL MSA PREDICTION OF THIS 2173 01:36:21,616 --> 01:36:24,185 PROTEIN WITH THE REDUCED TWO 2174 01:36:24,185 --> 01:36:26,854 SEQUENCE MSA THAT'S 70% ALANINE? 2175 01:36:26,854 --> 01:36:28,756 TURNS OUT THEY ARE COMPARABLE. 2176 01:36:28,756 --> 01:36:32,560 SO, YOU CAN SEE HERE THE 2177 01:36:32,560 --> 01:36:33,861 STRUCTURES ARE WITHIN ONE 2178 01:36:33,861 --> 01:36:35,963 ANGSTROM OF EACH OTHER, PLDTs 2179 01:36:35,963 --> 01:36:38,533 ARE PRETTY MUCH THE SAME SO IT 2180 01:36:38,533 --> 01:36:39,967 DOES SEEM THIS APPROACH IS 2181 01:36:39,967 --> 01:36:42,737 IDENTIFYING AT LEAST IN SOME 2182 01:36:42,737 --> 01:36:49,010 CASES INFORMATION THAT ALPHAFOD 2183 01:36:49,010 --> 01:36:49,644 IS USING. 2184 01:36:49,644 --> 01:36:51,713 WHERE ARE RESIDUES IN THE 2185 01:36:51,713 --> 01:36:51,946 PROTEIN? 2186 01:36:51,946 --> 01:36:53,014 THEY ARE ALL OVER THE PLACE. 2187 01:36:53,014 --> 01:36:54,982 SOME OF THEM ARE POINTING INWARD 2188 01:36:54,982 --> 01:36:57,952 AS WE WOULD EXPECT FOR CO- 2189 01:36:57,952 --> 01:36:59,020 EVOLUTIONARY INFORMATION, SOME 2190 01:36:59,020 --> 01:36:59,353 AREN'T. 2191 01:36:59,353 --> 01:37:00,955 I'VE DONE MORE STUDIES AND 2192 01:37:00,955 --> 01:37:02,290 ANALYSIS AND I CAN TALK TO YOU 2193 01:37:02,290 --> 01:37:03,558 MORE ABOUT THAT LATER. 2194 01:37:03,558 --> 01:37:05,293 BUT FOR NOW WHAT I WANT TO SAY 2195 01:37:05,293 --> 01:37:07,095 WHAT WE'RE TRYING TO DO IS IN 2196 01:37:07,095 --> 01:37:11,232 OUR NEXT STEPS APPLY TO HUNDRES 2197 01:37:11,232 --> 01:37:14,035 OF PROTEINS TO GET A SENSE OF 2198 01:37:14,035 --> 01:37:17,605 WHAT ALPHAFOLD HAS LEARNED TO 2199 01:37:17,605 --> 01:37:19,440 DEVELOP A MORE EFFICIENT MODEL 2200 01:37:19,440 --> 01:37:20,007 PREDICTING FOLD SWITCHING 2201 01:37:20,007 --> 01:37:21,943 PROTEINS MORE DIRECTLY. 2202 01:37:21,943 --> 01:37:23,244 IF WE UNDERSTAND PATTERNS 2203 01:37:23,244 --> 01:37:24,712 ALPHAFOLD IS USING TO PREDICT 2204 01:37:24,712 --> 01:37:27,582 CONFIRMATIONS WE CAN TEST THOSE 2205 01:37:27,582 --> 01:37:28,783 PATTERNS AGAINST GIVEN SEQUENCES 2206 01:37:28,783 --> 01:37:31,619 AND MAYBE BE ABLE TO ACHIEVE 2207 01:37:31,619 --> 01:37:32,386 ALTERNATIVE CONFIRMATION MORE 2208 01:37:32,386 --> 01:37:34,122 EFFICIENTLY THAN WE CAN NOW. 2209 01:37:34,122 --> 01:37:36,657 IN THE MEANTIME WE STILL HAVE CF 2210 01:37:36,657 --> 01:37:38,659 RANDOM WHICH WE HAVE RUN -- 2211 01:37:38,659 --> 01:37:39,494 >> QUICK QUESTION. 2212 01:37:39,494 --> 01:37:41,562 TIME CHECK HERE. 2213 01:37:41,562 --> 01:37:44,198 I SEE YOU'RE ON 50 OF 81 SLIDES, 2214 01:37:44,198 --> 01:37:46,901 THE SCHEDULE HAS YOU FINISHING 2215 01:37:46,901 --> 01:37:47,168 NOW. 2216 01:37:47,168 --> 01:37:48,136 SO -- 2217 01:37:48,136 --> 01:37:51,539 >> I'M NOT ACTUALLY -- I'M GOING 2218 01:37:51,539 --> 01:37:53,307 TO 60. 2219 01:37:53,307 --> 01:37:53,574 >> OKAY. 2220 01:37:53,574 --> 01:37:58,446 WE'RE GETTING CLOSE. 2221 01:37:58,446 --> 01:38:01,048 ALL RIGHT. 2222 01:38:01,048 --> 01:38:04,185 SO WE RAN CF RANDOM ON 2,000 E. 2223 01:38:04,185 --> 01:38:06,954 COLI PROTEINS AND GOT 2224 01:38:06,954 --> 01:38:08,823 PREDICTIONS OF ALTERNATIVE 2225 01:38:08,823 --> 01:38:10,391 CONFIRMATIONS FROM THOSE 2,000. 2226 01:38:10,391 --> 01:38:13,027 AND WE LOOKED FOR EVIDENCE OF 2227 01:38:13,027 --> 01:38:15,196 EVOLUTIONARY SELECTION OF THOSE 2228 01:38:15,196 --> 01:38:15,863 PROTEINS. 2229 01:38:15,863 --> 01:38:18,766 AND FROM THAT WE PREDICT UP TO 2230 01:38:18,766 --> 01:38:21,769 52 PROTEINS FROM THE E. COLI 2231 01:38:21,769 --> 01:38:23,004 GENOMES WHICH FOLDS, THEY HAVE 2232 01:38:23,004 --> 01:38:25,039 FUNCTION IN LINE WITH WHAT WE 2233 01:38:25,039 --> 01:38:27,875 SEE OF FOLD SWITCHING PROTEINS 2234 01:38:27,875 --> 01:38:29,610 SO TRANSCRIPTION TRANSLATION 2235 01:38:29,610 --> 01:38:30,611 REGULATOR IS SOMETHING WE 2236 01:38:30,611 --> 01:38:32,480 ALREADY KNOW IS ENRICHED IN FOLD 2237 01:38:32,480 --> 01:38:33,881 SWITCHING PROTEINS, FROM THIS WE 2238 01:38:33,881 --> 01:38:36,951 FIND -- WE ESTIMATE UP TO 5% OF 2239 01:38:36,951 --> 01:38:39,821 E. COLI PROTEINS SWITCH FOLDS 2240 01:38:39,821 --> 01:38:42,089 WHICH WE ARE WORKING WITH BRIAN 2241 01:38:42,089 --> 01:38:44,192 AND LOUIS TO TEST ALL OF THESE 2242 01:38:44,192 --> 01:38:45,326 PREDICTIONS EXPERIMENTALLY. 2243 01:38:45,326 --> 01:38:47,495 IN THE MEANTIME WHAT WE NEED ARE 2244 01:38:47,495 --> 01:38:50,064 EXPERIMENTAL METHODS TO BE ABLE 2245 01:38:50,064 --> 01:38:52,934 TO TEST FOR FOLD SWITCHING AT 2246 01:38:52,934 --> 01:38:53,167 SCALE. 2247 01:38:53,167 --> 01:38:55,236 LESLIE IN MY LAB IN 2248 01:38:55,236 --> 01:38:56,571 COLLABORATION WITH CAITLIN DAVIS 2249 01:38:56,571 --> 01:38:59,273 HAS DEVELOPED A FRET BASED ASSAY 2250 01:38:59,273 --> 01:39:00,441 FOR PROTEINS WITH DIFFERENT 2251 01:39:00,441 --> 01:39:02,777 PREDICTED END TO END DISTANCES, 2252 01:39:02,777 --> 01:39:05,980 WE'VE TESTED ON RFAH PROTEINS 2253 01:39:05,980 --> 01:39:07,381 AND PROTEINS WE CHARACTERIZED IN 2254 01:39:07,381 --> 01:39:08,249 THE LAB. 2255 01:39:08,249 --> 01:39:11,152 IT WORKS 13 OUT OF 15 TIMES AND 2256 01:39:11,152 --> 01:39:12,520 THE GREAT THING IS IT'S IN 2257 01:39:12,520 --> 01:39:15,256 CELLS, DOES NOT REQUIRE ANY 2258 01:39:15,256 --> 01:39:15,923 PROTEIN PURIFICATION. 2259 01:39:15,923 --> 01:39:19,026 SO WE WANT TO USE THE CELL AT 2260 01:39:19,026 --> 01:39:20,761 LARGE SCALE, LESLIE IS WORKING 2261 01:39:20,761 --> 01:39:22,630 ON THAT FOR RFAH AND THEN EXTEND 2262 01:39:22,630 --> 01:39:25,166 TO OTHER SYSTEMS. 2263 01:39:25,166 --> 01:39:27,134 WHAT WE WANT TO UNDERSTAND HOW 2264 01:39:27,134 --> 01:39:28,236 FOLD SWITCHING WORKS. 2265 01:39:28,236 --> 01:39:31,205 THAT MAY INFORM US TO BE ABLE TO 2266 01:39:31,205 --> 01:39:33,441 MAKE BETTER PREDICTIONS, ALSO 2267 01:39:33,441 --> 01:39:34,275 UNDERSTANDING MECHANISM IS 2268 01:39:34,275 --> 01:39:35,309 IMPORTANT FOR PROTEINS IN 2269 01:39:35,309 --> 01:39:36,177 GENERAL. 2270 01:39:36,177 --> 01:39:38,613 SO WHAT WE WANTED TO DO IS WE 2271 01:39:38,613 --> 01:39:41,582 HAVE THIS C-TERMINAL DOMAIN OF 2272 01:39:41,582 --> 01:39:42,950 RFAH WHICH IS HELICAL, BETA 2273 01:39:42,950 --> 01:39:44,252 SHEET IN ISOLATION. 2274 01:39:44,252 --> 01:39:49,523 WHAT WE WANT TO DO TO TITRATE 2275 01:39:49,523 --> 01:39:52,460 IN, FORCE IT TO ASSUME HELLICAL 2276 01:39:52,460 --> 01:39:54,428 CONFIRMATION BUT THE TECHNICAL 2277 01:39:54,428 --> 01:39:58,599 CHALLENGE IS THE NATIVE 2278 01:39:58,599 --> 01:40:00,434 N-TERMINAL DOMAIN OF RFAH IS 2279 01:40:00,434 --> 01:40:04,071 SOLUBLE TO 2 MICROMOLAR AND WE 2280 01:40:04,071 --> 01:40:05,172 END UP WITH AGGREGATION 2281 01:40:05,172 --> 01:40:05,439 PROGRESS. 2282 01:40:05,439 --> 01:40:09,043 WE NEED AT LEAST HUNDREDS OF 2283 01:40:09,043 --> 01:40:11,178 MICRO MOLARS TO DO THESE 2284 01:40:11,178 --> 01:40:11,479 EXPERIMENTS. 2285 01:40:11,479 --> 01:40:13,314 JOELY IN MY LAB WILL TELL YOU 2286 01:40:13,314 --> 01:40:13,714 MORE. 2287 01:40:13,714 --> 01:40:16,951 HE HAS BEEN ABLE TO DESIGN A 2288 01:40:16,951 --> 01:40:19,587 SOLUBLE N-TERMINAL DOMAIN, UP TO 2289 01:40:19,587 --> 01:40:22,423 AT LEAST 500 MICROMOLAR, 2290 01:40:22,423 --> 01:40:22,757 POSSIBLY MORE. 2291 01:40:22,757 --> 01:40:25,526 AND WE CAN SEE WHEN WE LOOK AT 2292 01:40:25,526 --> 01:40:26,727 THE ISOLATED C-TERMINAL DOMAIN 2293 01:40:26,727 --> 01:40:30,231 IT HAS ONE CHEMICAL SHIFT 2294 01:40:30,231 --> 01:40:31,599 PROFILE BY NMR. 2295 01:40:31,599 --> 01:40:38,873 IF WE ADD IN UNLABELED 2296 01:40:38,873 --> 01:40:41,075 N-TERMINAL DOMAIN WE SEE A 2297 01:40:41,075 --> 01:40:43,377 DRAMATIC SHIFT IN THE CHEMICAL 2298 01:40:43,377 --> 01:40:45,613 SHIFT PROFILE AND NATHAN IN THE 2299 01:40:45,613 --> 01:40:47,915 LAB IS WORKING ON ASSIGNING THIS 2300 01:40:47,915 --> 01:40:48,382 PROTEIN. 2301 01:40:48,382 --> 01:40:51,285 YOU CAN SEE WE'RE SEEING HELICAL 2302 01:40:51,285 --> 01:40:51,752 CHARACTER. 2303 01:40:51,752 --> 01:40:53,788 DOES SEEM THIS PROTEIN IS 2304 01:40:53,788 --> 01:40:54,422 WORKING. 2305 01:40:54,422 --> 01:40:59,126 THIS SYSTEM IS WORKING AS 2306 01:40:59,126 --> 01:40:59,927 EXPECTED AND COLLABORATING TO 2307 01:40:59,927 --> 01:41:01,762 TAKE SNAPSHOTS OF THE FOLD 2308 01:41:01,762 --> 01:41:03,898 SWITCHING PROCESS AS MUCH AS WE 2309 01:41:03,898 --> 01:41:04,065 CAN. 2310 01:41:04,065 --> 01:41:06,600 SO WITH THAT I'D LIKE TO THANK 2311 01:41:06,600 --> 01:41:11,005 MY LAB MEMBERS AND MY 2312 01:41:11,005 --> 01:41:12,206 COLLABORATORS, AND I'M HAPPY TO 2313 01:41:12,206 --> 01:41:12,907 TAKE QUESTIONS. 2314 01:41:12,907 --> 01:41:16,344 I APOLOGIZE FOR GOING OVER. 2315 01:41:16,344 --> 01:41:18,913 IT WAS HARD WITH THE TEAMS TO BE 2316 01:41:18,913 --> 01:41:20,614 ABLE TO ADVANCE MY SLIDES. 2317 01:41:20,614 --> 01:41:23,417 THANK YOU FOR YOUR PATIENCE WITH 2318 01:41:23,417 --> 01:41:23,918 THAT. 2319 01:41:23,918 --> 01:41:25,152 >> OKAY. 2320 01:41:25,152 --> 01:41:28,022 I THINK WE CAN RESET THE SYSTEM 2321 01:41:28,022 --> 01:41:32,560 HERE SO THAT WE DON'T HAVE -- WE 2322 01:41:32,560 --> 01:41:33,894 CAN DISPLAY PROPERLY ON THE 2323 01:41:33,894 --> 01:41:41,102 MONITORS IN THE ROOM HERE. 2324 01:41:41,102 --> 01:41:44,271 AUDIO/VISUAL GUY CAN BE HERE 2325 01:41:44,271 --> 01:41:54,682 MOMENTARILY, WE MIGHT DROP YOU 2326 01:41:54,682 --> 01:41:57,318 FOR A MOMENT. 2327 01:41:57,318 --> 01:42:07,395 . 2328 01:42:13,167 --> 01:42:16,103 >> DO YOU WANT TO STOP SHARING? 2329 01:42:16,103 --> 01:42:17,304 >> OKAY. 2330 01:42:17,304 --> 01:42:18,939 I'LL BE BACK. 2331 01:42:18,939 --> 01:42:20,241 >> STOP SHARING. 2332 01:42:20,241 --> 01:42:21,409 >> OH, STOP SHARING. 2333 01:42:21,409 --> 01:42:24,745 >> WE HAVE SOME QUESTIONS, 2334 01:42:24,745 --> 01:42:26,847 LAUREN, FROM THE BOARD. 2335 01:42:26,847 --> 01:42:34,955 BONNIE AND ANDRE A, GO AHEAD. 2336 01:42:34,955 --> 01:42:37,458 >> WE HAD TO RESET THE KODECK IN 2337 01:42:37,458 --> 01:42:38,559 THE ROOM SO IT'S GOING TO BE 2338 01:42:38,559 --> 01:42:39,493 JUST A MOMENT BEFORE THEY ARE 2339 01:42:39,493 --> 01:42:42,863 GOING TO BE ABLE TO HEAR YOU. 2340 01:42:42,863 --> 01:42:46,500 IF YOU COULD GIVE US A VERY 2341 01:42:46,500 --> 01:42:49,136 BRIEF PAUSE, WE'LL BE BACK 2342 01:42:49,136 --> 01:42:55,409 ONLINE IN JUST A MOMENT. 2343 01:42:55,409 --> 01:42:55,943 THANKS. 2344 01:42:55,943 --> 01:42:57,578 LOOKS LIKE THEY ARE BACK. 2345 01:42:57,578 --> 01:42:59,313 >> LOOKS LIKE LAUREN WAS STILL 2346 01:42:59,313 --> 01:43:01,315 THERE IN THE ROOM. 2347 01:43:01,315 --> 01:43:03,217 NOW SHE'S GONE. 2348 01:43:03,217 --> 01:43:04,485 >> I'M GONE NOW? 2349 01:43:04,485 --> 01:43:06,687 >> NO, SHE'S THERE. 2350 01:43:06,687 --> 01:43:07,054 >> WHERE? 2351 01:43:07,054 --> 01:43:11,592 NOW I CAN SEE HER. 2352 01:43:11,592 --> 01:43:14,128 I CAN SEE YOU, LAUREN. 2353 01:43:14,128 --> 01:43:15,996 THEN THERE ARE QUESTIONS FROM 2354 01:43:15,996 --> 01:43:21,535 BONNIE AND ANDREA. 2355 01:43:21,535 --> 01:43:23,537 GO AHEAD. 2356 01:43:23,537 --> 01:43:25,773 >> I WAS WONDERING HOW CF RANDOM 2357 01:43:25,773 --> 01:43:26,040 WORKED. 2358 01:43:26,040 --> 01:43:27,842 >> WHY DO I THINK IT WORKS? 2359 01:43:27,842 --> 01:43:29,143 >> NO, I DON'T EVEN KNOW WHAT 2360 01:43:29,143 --> 01:43:30,444 THE ALGORITHM IS. 2361 01:43:30,444 --> 01:43:36,383 WHAT YOU'RE USING FOR CF RANDOM. 2362 01:43:36,383 --> 01:43:36,650 >> SURE. 2363 01:43:36,650 --> 01:43:39,787 LET ME TRY TO SHARE MY SLIDES 2364 01:43:39,787 --> 01:43:40,221 AGAIN. 2365 01:43:40,221 --> 01:43:44,525 YEAH, I HAD TO BREEZE PAST THAT. 2366 01:43:44,525 --> 01:43:47,561 SO, LET ME GO BACK AND SHOW YOU. 2367 01:43:47,561 --> 01:43:51,398 THIS IS NOT THE MOST UPDATED BUT 2368 01:43:51,398 --> 01:43:54,802 IT DOES WELL ENOUGH. 2369 01:43:54,802 --> 01:43:56,070 SO WHAT WE'RE DOING -- 2370 01:43:56,070 --> 01:43:59,073 >> BEAUTIFUL WORK, I SHOULD HAVE 2371 01:43:59,073 --> 01:44:00,040 SAID. 2372 01:44:00,040 --> 01:44:01,976 >> THANK YOU. 2373 01:44:01,976 --> 01:44:08,048 SO, WE HAVE -- WE'RE USING -- 2374 01:44:08,048 --> 01:44:10,584 WHAT WE HAVE, WE'RE INPUTTING 2375 01:44:10,584 --> 01:44:13,120 JUST OUR INPUT SEQUENCE, AND A 2376 01:44:13,120 --> 01:44:16,590 FULL LENGTH MULTIPLE SEQUENCE 2377 01:44:16,590 --> 01:44:21,629 ALIGNMENT INTO THAT. 2378 01:44:21,629 --> 01:44:23,597 THAT GIVES DOMINANT CON 2379 01:44:23,597 --> 01:44:23,864 FIRMATION. 2380 01:44:23,864 --> 01:44:26,200 THE PROBLEM IS GETTING 2381 01:44:26,200 --> 01:44:28,402 ALTERNATIVE CONFIRMATION THAT'S 2382 01:44:28,402 --> 01:44:28,602 HARD. 2383 01:44:28,602 --> 01:44:34,642 WHAT WE'RE DOING IS SAMPLING 2384 01:44:34,642 --> 01:44:41,115 RANDOMLY GIVING OTHER PASSES 2385 01:44:41,115 --> 01:44:42,850 EXTREMELY SHALLOW MSAs, Seq 2386 01:44:42,850 --> 01:44:50,024 1 AND SEQ2, AND WE THINK WE'RE 2387 01:44:50,024 --> 01:44:51,659 DOING IS JOGGING ALPHAFOLD'S 2388 01:44:51,659 --> 01:44:55,729 MEMORY TO SEE HOW IT'S MAKING 2389 01:44:55,729 --> 01:45:00,100 ASSOCIATIONS WITH RANDOMLY 2390 01:45:00,100 --> 01:45:00,534 SHALLOW ALIGNMENTS. 2391 01:45:00,534 --> 01:45:02,636 LET ME SHOW WHAT YOU WE USED ON 2392 01:45:02,636 --> 01:45:03,704 THE E. COLI GENOME. 2393 01:45:03,704 --> 01:45:06,574 THIS IS WHY I HAVE THESE 2394 01:45:06,574 --> 01:45:06,807 BACKUPS. 2395 01:45:06,807 --> 01:45:07,908 SHOOT, DO I NOT HAVE IT? 2396 01:45:07,908 --> 01:45:09,176 HERE IT IS. 2397 01:45:09,176 --> 01:45:09,476 OKAY. 2398 01:45:09,476 --> 01:45:11,045 HERE WE GO. 2399 01:45:11,045 --> 01:45:13,214 WE'VE GOT OUR DOMINANT STRUCTURE 2400 01:45:13,214 --> 01:45:15,249 SEARCH, FULL MSA, AND OUR 2401 01:45:15,249 --> 01:45:15,583 SHALLOW MSA. 2402 01:45:15,583 --> 01:45:17,218 WE'RE CLEAR ON THAT. 2403 01:45:17,218 --> 01:45:20,521 NOW WE NEED AN UNBIASED WAY TO 2404 01:45:20,521 --> 01:45:21,121 PREDICT ALTERNATIVE 2405 01:45:21,121 --> 01:45:21,889 CONFIRMATIONS BECAUSE HOW ALSO 2406 01:45:21,889 --> 01:45:24,658 ARE WE GOING TO RUN IT? 2407 01:45:24,658 --> 01:45:25,759 >> PEOPLE ARE HAVING TROUBLE 2408 01:45:25,759 --> 01:45:27,761 SEEING YOUR SLIDES. 2409 01:45:27,761 --> 01:45:33,400 MONA, I WOULD GO THROUGH LIKE A 2410 01:45:33,400 --> 01:45:34,602 DIFFERENT BROWSER. 2411 01:45:34,602 --> 01:45:39,940 LIKE SAFARI, PASTE THE LINK INTO 2412 01:45:39,940 --> 01:45:40,274 GOOGLE CHROME. 2413 01:45:40,274 --> 01:45:43,344 OR SOMETHING. 2414 01:45:43,344 --> 01:45:45,846 I DON'T THINK IT WORKS THROUGH 2415 01:45:45,846 --> 01:45:47,047 TEAMS DIRECTLY. 2416 01:45:47,047 --> 01:45:48,148 >> I'M ACTUALLY GETTING 2417 01:45:48,148 --> 01:45:51,885 EVERYTHING WORKING NOW THROUGH 2418 01:45:51,885 --> 01:45:53,520 TEAMS DIRECTLY BUT THE AV FOLKS 2419 01:45:53,520 --> 01:45:56,156 SAID GO TO VIEW AND GALLERY VIEW 2420 01:45:56,156 --> 01:45:58,359 AT TOP, YOU SEE PEOPLE UP AT THE 2421 01:45:58,359 --> 01:46:06,533 TOP, THE SLIDE DOWN AT THE 2422 01:46:06,533 --> 01:46:06,767 BOTTOM. 2423 01:46:06,767 --> 01:46:07,735 RICHARD'S SEEING IT. 2424 01:46:07,735 --> 01:46:13,741 >> AND I'M SEEING IT ON TEAMS. 2425 01:46:13,741 --> 01:46:16,944 SO A VIEW WITH NO SLIDES FOR 2426 01:46:16,944 --> 01:46:18,912 SOME PEOPLE. 2427 01:46:18,912 --> 01:46:20,314 MONA IF YOU -- 2428 01:46:20,314 --> 01:46:25,853 >> YOU'RE CHANGING THE RECYCLING 2429 01:46:25,853 --> 01:46:30,491 PROCESS BY INPUTTING SMALLER 2430 01:46:30,491 --> 01:46:31,692 MSAs, AND THEREFORE IT'S 2431 01:46:31,692 --> 01:46:33,594 FASTER, DOESN'T HAVE TO DO AS 2432 01:46:33,594 --> 01:46:35,863 MUCH WORK, AND YOU'RE 2433 01:46:35,863 --> 01:46:36,864 POTENTIALLY CAPTURING THE 2434 01:46:36,864 --> 01:46:41,268 SEQUENCES THAT LEAD TO THESE 2435 01:46:41,268 --> 01:46:44,305 OTHER STRUCTURES BETTER. 2436 01:46:44,305 --> 01:46:45,939 SO WE'RE NOT -- I DON'T THINK 2437 01:46:45,939 --> 01:46:48,509 WE'RE NECESSARILY -- DO YOU SEE 2438 01:46:48,509 --> 01:46:50,577 MY SLIDE NOW? 2439 01:46:50,577 --> 01:46:52,513 >> I'M SEEING YOUR SLIDE. 2440 01:46:52,513 --> 01:46:55,716 I TRYING TO HELP OTHERS SEE YOUR 2441 01:46:55,716 --> 01:46:56,750 SLIDE. 2442 01:46:56,750 --> 01:46:57,418 >> SORRY. 2443 01:46:57,418 --> 01:47:00,487 SO WE'RE USING ALPHAFOLD'S 2444 01:47:00,487 --> 01:47:04,892 RECYCLING AS THEY USE IT, A 2445 01:47:04,892 --> 01:47:08,862 DIFFERENT TINY SHALLOW RANDOM 2446 01:47:08,862 --> 01:47:12,099 RSA, AND RANDOMLY SELECTED. 2447 01:47:12,099 --> 01:47:14,535 >> HAVE YOU TRIED MODELS THAT 2448 01:47:14,535 --> 01:47:18,305 ARE JUST PROTEIN LANGUAGE MODELS 2449 01:47:18,305 --> 01:47:21,875 THAT DON'T NEED ALIGNMENT. 2450 01:47:21,875 --> 01:47:22,176 >> YES. 2451 01:47:22,176 --> 01:47:24,278 WE HAVE TRIED THOSE MODELS AND 2452 01:47:24,278 --> 01:47:27,047 WHAT WE'VE FOUND IS THAT 2453 01:47:27,047 --> 01:47:29,183 ESPECIALLY ON SWITCHING PROTEINS 2454 01:47:29,183 --> 01:47:33,554 THEY TENDING TO OVERTRAINED. 2455 01:47:33,554 --> 01:47:35,689 EVEN MORE SO THAN ALPHAFOLD, 2456 01:47:35,689 --> 01:47:44,298 ALTERNATIVE CON CONFIRMATION. 2457 01:47:44,298 --> 01:47:46,066 IF WE MASK OUT C-TERMINAL 2458 01:47:46,066 --> 01:47:48,702 DOMAINS IT DOESN'T SEE 2459 01:47:48,702 --> 01:47:49,536 SEQUENCES, GIVES HELICAL 2460 01:47:49,536 --> 01:47:49,870 CONFIRMATION. 2461 01:47:49,870 --> 01:47:52,439 WE DID NOT WANT TO GO THAT WAY 2462 01:47:52,439 --> 01:47:55,109 WHEN DOING THESE PREDICTIONS. 2463 01:47:55,109 --> 01:47:57,711 >> TRAINED IN A SENSE THAT GAVE 2464 01:47:57,711 --> 01:47:58,712 THE ALTERNATIVE CONFIRMATIONS, 2465 01:47:58,712 --> 01:48:02,182 YOU THOUGHT THEY WERE 2466 01:48:02,182 --> 01:48:03,384 OVERTRAINED ON THOSE? 2467 01:48:03,384 --> 01:48:05,119 >> OVER TRAINED WHEN WE GAVE IT 2468 01:48:05,119 --> 01:48:08,589 NO INFORMATION IT WAS GIVING US 2469 01:48:08,589 --> 01:48:09,223 ALTERNATIVE CONFIRMATION. 2470 01:48:09,223 --> 01:48:09,857 >> WAIT A MINUTE. 2471 01:48:09,857 --> 01:48:12,159 BUT THAT'S A GOOD THING. 2472 01:48:12,159 --> 01:48:14,928 YOU DON'T KNOW THAT THEY ARE 2473 01:48:14,928 --> 01:48:15,462 OVERTRAINED. 2474 01:48:15,462 --> 01:48:17,631 >> NO, I DO THINK IT'S 2475 01:48:17,631 --> 01:48:19,233 OVERTRAINED BECAUSE IF WE MASK 2476 01:48:19,233 --> 01:48:22,603 THE WHOLE SEQUENCE OF THE 2477 01:48:22,603 --> 01:48:24,004 C-TERMINAL DOMAIN, ALMOST ALL OF 2478 01:48:24,004 --> 01:48:25,372 IT, WE'RE GETTING THE RIGHT 2479 01:48:25,372 --> 01:48:25,973 ANSWER. 2480 01:48:25,973 --> 01:48:27,541 I DON'T THINK THAT'S A GOOD 2481 01:48:27,541 --> 01:48:28,142 THING. 2482 01:48:28,142 --> 01:48:29,143 THAT'S NOT ROBUST. 2483 01:48:29,143 --> 01:48:33,414 I HAVE OTHER WORK SHOWING 2484 01:48:33,414 --> 01:48:35,149 ALPHAFOLD ALSO IS WAY 2485 01:48:35,149 --> 01:48:38,819 OVERPREDICTING ALTERNATIVE 2486 01:48:38,819 --> 01:48:39,520 CONFIRMS, ONLY REQUIRING TWO 2487 01:48:39,520 --> 01:48:42,556 AMINO ACIDS TO MAKE PREDICTIONS, 2488 01:48:42,556 --> 01:48:43,190 WAY OVERPREDICTING ALTERNATIVE 2489 01:48:43,190 --> 01:48:44,258 CON FIRMATION. 2490 01:48:44,258 --> 01:48:47,227 HERE IS WHAT I'M TALKING ABOUT. 2491 01:48:47,227 --> 01:48:50,597 THIS IS A PROTEIN XCL 1, TO GET 2492 01:48:50,597 --> 01:48:55,202 IT TO PREDICT ALTERNATIVE CON 2493 01:48:55,202 --> 01:48:56,637 FIRMATION FROM SINGLE SEQUENCE, 2494 01:48:56,637 --> 01:48:58,305 EXPERIMENTAL TEST THE VARIANTS. 2495 01:48:58,305 --> 01:49:01,308 IF WE ONLY CHANGE TWO AMINO 2496 01:49:01,308 --> 01:49:04,978 ACIDS, WE GET IT TO ALWAYS 2497 01:49:04,978 --> 01:49:06,246 PREDICT THAT ALTERNATIVE 2498 01:49:06,246 --> 01:49:07,381 CONFIRMATION BUT THAT'S NOT NOW 2499 01:49:07,381 --> 01:49:08,916 IT'S WORKING. 2500 01:49:08,916 --> 01:49:12,886 IT'S NOT ROBUST FOR THIS 2501 01:49:12,886 --> 01:49:13,320 PROBLEM. 2502 01:49:13,320 --> 01:49:20,494 >> I'LL LET OTHERS ASK THEIR 2503 01:49:20,494 --> 01:49:20,761 QUESTIONS 2504 01:49:20,761 --> 01:49:25,332 >> IT'S MY TURN, RIGHT? 2505 01:49:25,332 --> 01:49:29,169 FANTASTIC TALK. 2506 01:49:29,169 --> 01:49:29,603 REALLY ILLUMINATING. 2507 01:49:29,603 --> 01:49:35,742 BUT FIRST, WHY YOU NEED TWO 2508 01:49:35,742 --> 01:49:37,077 DIFFERENT ESSENTIALLY FUNCTIONS 2509 01:49:37,077 --> 01:49:39,213 OF THE SAME PROTEIN, RIGHT? 2510 01:49:39,213 --> 01:49:40,647 WHAT'S BIOLOGIC NECESSITY OF 2511 01:49:40,647 --> 01:49:44,384 HAVING TWO DIFFERENT STRUCTURES, 2512 01:49:44,384 --> 01:49:44,585 FIRST. 2513 01:49:44,585 --> 01:49:45,919 SECOND, DID I UNDERSTAND 2514 01:49:45,919 --> 01:49:48,155 CORRECTLY YOU CAN DETECT 2515 01:49:48,155 --> 01:49:51,725 CO-EVOLUTIONARY CHANGES IN BOTH 2516 01:49:51,725 --> 01:49:52,292 STRUCTURES, RIGHT? 2517 01:49:52,292 --> 01:49:52,793 >> OKAY. 2518 01:49:52,793 --> 01:49:54,461 YES, SO TO ANSWER YOUR FIRST 2519 01:49:54,461 --> 01:49:57,030 QUESTION THERE ARE A FEW REASONS 2520 01:49:57,030 --> 01:49:58,031 FOLD SWITCHING EXISTS. 2521 01:49:58,031 --> 01:50:02,603 ONE IS BECAUSE IT CAN BE A 2522 01:50:02,603 --> 01:50:03,804 REGULATORY MECHANISM. 2523 01:50:03,804 --> 01:50:05,639 SO THAT YOU CAN IMAGINE CASES 2524 01:50:05,639 --> 01:50:07,407 LIKE FOR CHI B IT'S IN ONE 2525 01:50:07,407 --> 01:50:09,943 STATE, MOST OF THE TIME BECAUSE 2526 01:50:09,943 --> 01:50:11,211 IT'S SWITCHING VERY SLOWLY THAT 2527 01:50:11,211 --> 01:50:12,946 ALLOWS THE CLOCK TO HAVE THE 2528 01:50:12,946 --> 01:50:13,647 RIGHT TIMING. 2529 01:50:13,647 --> 01:50:15,782 OR YOU CAN IMAGINE A PROTEIN 2530 01:50:15,782 --> 01:50:18,585 BEING ABLE TO RESPOND QUICKLY IN 2531 01:50:18,585 --> 01:50:19,686 RESPONSE TO CELLULAR STIMULI. 2532 01:50:19,686 --> 01:50:21,421 SO USUALLY WHEN PEOPLE ARE 2533 01:50:21,421 --> 01:50:24,091 THINKING ABOUT THAT, THEY THINK 2534 01:50:24,091 --> 01:50:25,225 ABOUT SUBFUNCTIONALIZATION, 2535 01:50:25,225 --> 01:50:28,362 ANOTHER PROTEIN COPIES AND THEN 2536 01:50:28,362 --> 01:50:29,930 IT RESPONDS TO CELLULAR CHANGES 2537 01:50:29,930 --> 01:50:31,131 BY PRODUCING ANOTHER PROTEIN. 2538 01:50:31,131 --> 01:50:33,934 BUT IT'S MORE COST EFFECTIVE TO 2539 01:50:33,934 --> 01:50:35,102 JUST HAVE A PROTEIN SWITCH BE 2540 01:50:35,102 --> 01:50:35,969 ABLE TO DO SOMETHING. 2541 01:50:35,969 --> 01:50:37,304 I THINK THERE ARE JUST CASES 2542 01:50:37,304 --> 01:50:40,574 WHERE IT MAKES SENSE TO HAVE 2543 01:50:40,574 --> 01:50:41,108 ENVIRONMENTALLY RESPONSIVE 2544 01:50:41,108 --> 01:50:46,079 PROTEINS IN THE CELL THAT HAVE 2545 01:50:46,079 --> 01:50:46,480 COUPLED FUNCTIONS. 2546 01:50:46,480 --> 01:50:47,881 TO ANSWER THE QUESTION OF 2547 01:50:47,881 --> 01:50:50,284 EVOLUTIONARY SELECTION WE DID A 2548 01:50:50,284 --> 01:50:52,085 STUDY IN 2023, IT WAS PUBLISHED, 2549 01:50:52,085 --> 01:50:57,124 WE WERE ABLE TO SEE SIGNS OF 2550 01:50:57,124 --> 01:50:59,426 DUAL FOLD CO-EVOLUTION, IN 58 I 2551 01:50:59,426 --> 01:51:02,062 THINK FOLD SWITCHING PROTEINS. 2552 01:51:02,062 --> 01:51:04,498 AND WE CAN ONLY DO IT BECAUSE WE 2553 01:51:04,498 --> 01:51:05,999 NEEDED DEEP MSAs TO DO IT BUT 2554 01:51:05,999 --> 01:51:07,968 IN THOSE CASES WE DID SEE IT. 2555 01:51:07,968 --> 01:51:11,204 I DO THINK THAT AT LEAST IN SOME 2556 01:51:11,204 --> 01:51:14,508 CASES FOLD SWITCHING IS SELECTED 2557 01:51:14,508 --> 01:51:16,510 BY EVOLUTION. 2558 01:51:16,510 --> 01:51:18,812 >> THANK YOU SO MUCH. 2559 01:51:18,812 --> 01:51:25,786 >> YEAH. 2560 01:51:25,786 --> 01:51:28,555 >> AM I NEXT? 2561 01:51:28,555 --> 01:51:29,823 THERE WE GO. 2562 01:51:29,823 --> 01:51:32,392 ANYWAY, FANTASTIC STUFF, REALLY 2563 01:51:32,392 --> 01:51:33,026 ILLUMINATING. 2564 01:51:33,026 --> 01:51:35,295 UNFORTUNATELY I HAD TO RELOG IN 2565 01:51:35,295 --> 01:51:37,731 WHEN BONNIE WAS ASKING THE 2566 01:51:37,731 --> 01:51:39,366 QUESTION ABOUT CF RANDOM, HOW 2567 01:51:39,366 --> 01:51:40,801 DID IT WORK. 2568 01:51:40,801 --> 01:51:43,503 LET ME PARAPHRASE IN TERMS OF 2569 01:51:43,503 --> 01:51:46,306 WHEN DO YOU THINK YOU'RE 2570 01:51:46,306 --> 01:51:48,175 LEARNING ADDITIONAL PHYSICS 2571 01:51:48,175 --> 01:51:49,543 THROUGH YOUR MODIFICATION OF THE 2572 01:51:49,543 --> 01:51:51,345 ALPHAFOLD ALGORITHM? 2573 01:51:51,345 --> 01:51:53,213 AND THAT'S WHAT'S HELPING YOU 2574 01:51:53,213 --> 01:51:54,214 FIND THE SECOND CONFIRMATION? 2575 01:51:54,214 --> 01:51:56,383 >> I DO NOT THINK WE'RE 2576 01:51:56,383 --> 01:52:01,121 LEARNING PHYSICS. 2577 01:52:01,121 --> 01:52:02,289 AND ACTUALLY DUBLINA WILL SHOW 2578 01:52:02,289 --> 01:52:05,392 YOU WHY IN HER TALK. 2579 01:52:05,392 --> 01:52:07,427 I THINK ALPHAFOLD IS REALLY GOOD 2580 01:52:07,427 --> 01:52:09,896 AT KNOWING WHAT A PLAUSIBLE 2581 01:52:09,896 --> 01:52:10,697 PROTEIN STRUCTURE LOOKS LIKE, 2582 01:52:10,697 --> 01:52:12,499 BUT I DON'T THINK THAT'S 2583 01:52:12,499 --> 01:52:12,966 PHYSICS. 2584 01:52:12,966 --> 01:52:17,070 IF IT'S OKAY WITH YOU I'LL LET 2585 01:52:17,070 --> 01:52:18,805 YOU SEE BECAUSE DUBLINA HAS HER 2586 01:52:18,805 --> 01:52:21,642 SLIDES AND WILL SHOW YOU WHY. 2587 01:52:21,642 --> 01:52:23,977 IS THAT OKAY OR DO YOU WANT A 2588 01:52:23,977 --> 01:52:24,411 MORE -- 2589 01:52:24,411 --> 01:52:26,446 >> JUST AS A TIME CHECK, WHY 2590 01:52:26,446 --> 01:52:28,148 DON'T YOU HAVE THE LAST QUESTION 2591 01:52:28,148 --> 01:52:35,222 SO WE DON'T CUT MUCH INTO THE 2592 01:52:35,222 --> 01:52:36,023 POSTDOCS' TIME. 2593 01:52:36,023 --> 01:52:36,723 >> LAST QUESTION FIRST -- 2594 01:52:36,723 --> 01:52:39,393 >> I CAN'T HEAR YOU. 2595 01:52:39,393 --> 01:52:42,729 SORRY, COULD YOU SAY THAT AGAIN? 2596 01:52:42,729 --> 01:52:44,398 >> YES, REALLY REMARKABLE WORK 2597 01:52:44,398 --> 01:52:47,467 THAT YOU'RE DOING. 2598 01:52:47,467 --> 01:52:48,001 PARTICULARLY COMBINING -- 2599 01:52:48,001 --> 01:52:53,540 (INAUDIBLE). 2600 01:52:53,540 --> 01:52:55,509 VERY SHORT QUESTIONS, SEEMS THE 2601 01:52:55,509 --> 01:52:58,178 FOLD SWITCHING COULD COME FROM 2602 01:52:58,178 --> 01:53:03,216 DIFFERENT REASONS, AND SOME 2603 01:53:03,216 --> 01:53:05,385 COULD BE A TYPE OF (INAUDIBLE) I 2604 01:53:05,385 --> 01:53:10,691 DON'T KNOW FOR THE ONE EXAMPLE 2605 01:53:10,691 --> 01:53:13,927 YOU'RE SHOWING US, MUTATION OR 2606 01:53:13,927 --> 01:53:14,261 ENVIRONMENTAL. 2607 01:53:14,261 --> 01:53:16,463 HOW DOES THE ORIGIN OF THE FOLD 2608 01:53:16,463 --> 01:53:22,703 SWITCHING CHANGE AS YOUR 2609 01:53:22,703 --> 01:53:24,404 ALGORITHMS MOVES TO EXPLORE AND 2610 01:53:24,404 --> 01:53:24,638 PREDICT? 2611 01:53:24,638 --> 01:53:26,173 >> THIS IS A GREAT QUESTION. 2612 01:53:26,173 --> 01:53:28,775 SO I THINK WHAT WE HAVE NOW, 2613 01:53:28,775 --> 01:53:31,511 LIKE REALLY THINK WHAT CF RANDOM 2614 01:53:31,511 --> 01:53:33,046 DOES MOSTLY IS LEVERAGE 2615 01:53:33,046 --> 01:53:34,548 STRUCTURES THAT ALPHAFOLD HAS 2616 01:53:34,548 --> 01:53:36,550 ALREADY SEEN DURING ITS TRAINING 2617 01:53:36,550 --> 01:53:39,186 AND ASSOCIATE THAT WITH AN INPUT 2618 01:53:39,186 --> 01:53:39,553 STRUCTURE. 2619 01:53:39,553 --> 01:53:42,255 AND I DON'T -- SO THAT'S USEFUL 2620 01:53:42,255 --> 01:53:43,457 TO AN EXTENT. 2621 01:53:43,457 --> 01:53:45,158 BUT IT'S NOT FULLY USEFUL. 2622 01:53:45,158 --> 01:53:46,526 SO ONE EXAMPLE WHERE I THINK 2623 01:53:46,526 --> 01:53:48,829 THIS IS WHY WE'RE DOING A 2624 01:53:48,829 --> 01:53:49,996 DETAILED INTERPRETATION OF 2625 01:53:49,996 --> 01:53:51,665 ALPHAFOLD TRYING TO UNDERSTAND 2626 01:53:51,665 --> 01:53:57,104 WHAT IT'S SEEING IS WITH THAT 2627 01:53:57,104 --> 01:53:59,172 BCCIP ALPHA PROTEIN, BETA IS 2628 01:53:59,172 --> 01:54:00,207 REALLY WELL CONSERVED, ALL THE 2629 01:54:00,207 --> 01:54:04,044 WAY DOWN TO YEAST. 2630 01:54:04,044 --> 01:54:05,579 AND HIGHER EUKARYOTES. 2631 01:54:05,579 --> 01:54:07,547 ALPHA IS CONSERVED IN PRIMATES 2632 01:54:07,547 --> 01:54:09,583 AND A FEW OTHER ORGANISMS. 2633 01:54:09,583 --> 01:54:12,486 THERE'S NOT THAT MUCH GENETIC 2634 01:54:12,486 --> 01:54:14,020 VARIATION AMONG HIGHER 2635 01:54:14,020 --> 01:54:14,588 EUKARYOTES. 2636 01:54:14,588 --> 01:54:17,257 TO SEE SOME SORT OF 2637 01:54:17,257 --> 01:54:19,593 CO-EVOLUTIONARY SIGNAL TO MAKE 2638 01:54:19,593 --> 01:54:20,994 PREDICTIONS MAY NOT BE POSSIBLE. 2639 01:54:20,994 --> 01:54:23,530 WE NEED OTHER APPROACHES TO 2640 01:54:23,530 --> 01:54:25,298 PREDICT ALTERNATIVE CON 2641 01:54:25,298 --> 01:54:25,599 FIRMATIONS. 2642 01:54:25,599 --> 01:54:26,867 ANOTHER CHALLENGE WITH THIS 2643 01:54:26,867 --> 01:54:28,802 RESEARCH YOU MAY BE ALLUDING TO 2644 01:54:28,802 --> 01:54:31,438 IS THAT WE DON'T KNOW WHAT 2645 01:54:31,438 --> 01:54:32,973 TRIGGERS FOLD SWITCHING. 2646 01:54:32,973 --> 01:54:35,575 AND THAT'S A REALLY BIG 2647 01:54:35,575 --> 01:54:35,842 CHALLENGE. 2648 01:54:35,842 --> 01:54:37,010 WITH EXPERIMENTS WHAT WE'RE 2649 01:54:37,010 --> 01:54:41,047 HOPING FOR IS TO RUN NMR 2650 01:54:41,047 --> 01:54:42,549 EXPERIMENTS OR FRET EXPERIMENTS, 2651 01:54:42,549 --> 01:54:44,251 DO WE SEE TWO OR DO WE SEE 2652 01:54:44,251 --> 01:54:49,055 ANYTHING, THIS IS WHY WE'RE 2653 01:54:49,055 --> 01:54:50,957 COLLABORATING WITH NUMEROUS NMR 2654 01:54:50,957 --> 01:54:51,191 PEOPLE. 2655 01:54:51,191 --> 01:54:55,162 WE'VE BEEN WORKING ON A HYDROGEN 2656 01:54:55,162 --> 01:54:56,863 EXCHANGE BASED SCREENING METHOD 2657 01:54:56,863 --> 01:55:02,836 WITH SUSAN'S LAB, MAYBE ALLOWIG 2658 01:55:02,836 --> 01:55:04,771 US TO OVERCOME SOME BARRIERS, 2659 01:55:04,771 --> 01:55:11,311 BUT KNOWING THE TRIGGER IS AN 2660 01:55:11,311 --> 01:55:12,479 ACHILLES HEEL AND WE'RE HOPING 2661 01:55:12,479 --> 01:55:13,613 TO PICK UP SOMETHING GOOD FOR 2662 01:55:13,613 --> 01:55:16,049 NOW BUT WE HOPE AS WE GET MORE 2663 01:55:16,049 --> 01:55:17,784 EXAMPLES AND UNDERSTAND THE 2664 01:55:17,784 --> 01:55:19,452 PRINCIPLES BETTER, WE MAY BE 2665 01:55:19,452 --> 01:55:23,657 ABLE TO DO THINGS IN A MORE 2666 01:55:23,657 --> 01:55:30,797 INFORMED WAY. 2667 01:55:30,797 --> 01:55:31,097 >> GREAT. 2668 01:55:31,097 --> 01:55:32,232 WE'LL HAVE -- THE BOARD WILL 2669 01:55:32,232 --> 01:55:35,414 HAVE A CHANCE TO ASK QUESTIONS 2670 01:55:35,414 --> 01:55:37,449 SO I WILL RECONVENE THE MEETING 2671 01:55:37,449 --> 01:55:38,417 OF THE NATIONAL LIBRARY OF 2672 01:55:38,417 --> 01:55:40,052 MEDICINE BOARD OF SCIENTIFIC 2673 01:55:40,052 --> 01:55:40,352 COUNSELORS. 2674 01:55:40,352 --> 01:55:42,554 WE'RE NOW IN PUBLIC SESSION, 2675 01:55:42,554 --> 01:55:45,057 RECORDING AND BROADCASTING. 2676 01:55:45,057 --> 01:55:48,627 PLEASE CONFIRM, AV, FOLKS, THAT 2677 01:55:48,627 --> 01:55:51,029 THAT'S THE CASE. 2678 01:55:51,029 --> 01:55:51,863 >> GOOD. 2679 01:55:51,863 --> 01:55:52,831 >> GREAT. 2680 01:55:52,831 --> 01:55:59,071 SO, JEREMY, GO AHEAD AND START. 2681 01:55:59,071 --> 01:56:01,406 THANK YOU FOR BEING HERE AND 2682 01:56:01,406 --> 01:56:03,075 PRESENTING TO US, JEREMY WEISS. 2683 01:56:03,075 --> 01:56:05,611 YOU HAVE ABOUT 30 MINUTES HERE. 2684 01:56:05,611 --> 01:56:07,079 >> ALL RIGHT. 2685 01:56:07,079 --> 01:56:09,748 CAN EVERYBODY SEE THE SCREEN 2686 01:56:09,748 --> 01:56:09,948 OKAY? 2687 01:56:09,948 --> 01:56:11,249 >> YES. 2688 01:56:11,249 --> 01:56:12,584 >> OKAY, PERFECT. 2689 01:56:12,584 --> 01:56:14,553 SO, WELCOME, EVERYBODY. 2690 01:56:14,553 --> 01:56:17,255 MY NAME IS JEREMY WEISS, AN 2691 01:56:17,255 --> 01:56:19,124 INVESTIGATOR COMPUTATIONAL 2692 01:56:19,124 --> 01:56:22,628 HEALTH AT THE NLM, JOINED IN 2693 01:56:22,628 --> 01:56:25,998 2022, WITH THE BOARD OF 2694 01:56:25,998 --> 01:56:27,332 SCIENTIFIC COUNSELORS. 2695 01:56:27,332 --> 01:56:29,968 TODAY'S TALK IS ENTITLED 2696 01:56:29,968 --> 01:56:34,272 QUANTITATIVE MACHINE LEARNING 2697 01:56:34,272 --> 01:56:37,309 FOR RISK CHARACTERIZATION. 2698 01:56:37,309 --> 01:56:40,545 I LEAD THE LAB, CARE HEALTH AND 2699 01:56:40,545 --> 01:56:42,381 REASONING, AND OUR RESEARCH 2700 01:56:42,381 --> 01:56:44,383 GOALS TO ADVANCE CAPABILITIES OF 2701 01:56:44,383 --> 01:56:46,618 TEMPORAL MACHINE LEARNING IN 2702 01:56:46,618 --> 01:56:48,186 CLINICAL SETTINGS, RISK 2703 01:56:48,186 --> 01:56:49,955 CHARACTERIZATION OF DISEASE, AND 2704 01:56:49,955 --> 01:56:51,757 MY GROUP IS HERE. 2705 01:56:51,757 --> 01:56:54,359 I HAVE THREE FELLOWS, ONE WILL 2706 01:56:54,359 --> 01:56:58,530 BE PRESENTING A POSTER, A 2707 01:56:58,530 --> 01:57:00,365 Ph.D. STUDENT, AND SHANTON AND 2708 01:57:00,365 --> 01:57:05,237 JING WANG WHO JOINED IN 2025 AND 2709 01:57:05,237 --> 01:57:07,139 2024 RESPECTIVELY. 2710 01:57:07,139 --> 01:57:08,707 IN TODAY'S TALK I'D LIKE TO 2711 01:57:08,707 --> 01:57:10,575 FIRST START WITH A MOTIVATION 2712 01:57:10,575 --> 01:57:13,278 FROM THE LENS OF CLINICAL 2713 01:57:13,278 --> 01:57:14,746 DECISION MAKING AND OVERLAYING 2714 01:57:14,746 --> 01:57:16,048 OUR THINKING ABOUT RISK 2715 01:57:16,048 --> 01:57:17,582 CHARACTERIZATION ON TOP OF THAT. 2716 01:57:17,582 --> 01:57:19,117 AND I'LL GIVE A HIGH LEVEL 2717 01:57:19,117 --> 01:57:22,754 OVERVIEW OF OUR GROUP'S WORK AND 2718 01:57:22,754 --> 01:57:24,289 LONGITUDINAL RISK PREDICTION 2719 01:57:24,289 --> 01:57:26,158 FOCUSING IN PARTICULAR ON 2720 01:57:26,158 --> 01:57:26,425 SUBGROUPS. 2721 01:57:26,425 --> 01:57:29,261 WHETHER THEY BE CLUSTERS OR RISK 2722 01:57:29,261 --> 01:57:31,263 STRATA OR LONGITUDINAL 2723 01:57:31,263 --> 01:57:32,597 TRAJECTORIES OR PATHWAYS. 2724 01:57:32,597 --> 01:57:38,670 I'LL DO A DEEPER DIVE INTO 2725 01:57:38,670 --> 01:57:38,970 RECENT WORK. 2726 01:57:38,970 --> 01:57:41,973 I'D LIKE TO INTRODUCE THE 2727 01:57:41,973 --> 01:57:43,341 CONCEPTUAL FRAMEWORK WHICH IS 2728 01:57:43,341 --> 01:57:46,912 TRYING TO GET AT HOW DO WE 2729 01:57:46,912 --> 01:57:49,948 TRANSLATE OUR BODY OF KNOWLEDGE 2730 01:57:49,948 --> 01:57:52,617 FROM BODY OF EVIDENCE INTO THE 2731 01:57:52,617 --> 01:57:53,418 DECISION-MAKING PROCESS AT THE 2732 01:57:53,418 --> 01:57:55,020 POINT OF CARE. 2733 01:57:55,020 --> 01:57:56,621 I'D ARGUE THERE'S REALLY TWO 2734 01:57:56,621 --> 01:57:59,858 PILLARS OF THIS PROCESS, FIRST 2735 01:57:59,858 --> 01:58:04,629 IS USE OF EVIDENCE-BASED 2736 01:58:04,629 --> 01:58:08,133 PROTOCOLS, AND THINKING ABOUT 2737 01:58:08,133 --> 01:58:09,234 CLINICAL EQUIPOISE AT THE POINT 2738 01:58:09,234 --> 01:58:10,435 OF CARE, DESCRIBED IN DETAIL, 2739 01:58:10,435 --> 01:58:11,737 LET'S LOOK AT THE DIAGRAM. 2740 01:58:11,737 --> 01:58:14,272 ON THE LEFT WE HAVE THE LEVELS 2741 01:58:14,272 --> 01:58:17,042 OF EVIDENCE FROM EVIDENCE-BASED 2742 01:58:17,042 --> 01:58:19,978 MEDICINE, THE TYPES OF DATA THAT 2743 01:58:19,978 --> 01:58:22,614 AND STUDIES WE CONDUCT TO 2744 01:58:22,614 --> 01:58:23,815 GENERATE OUR BODY OF KNOWLEDGE, 2745 01:58:23,815 --> 01:58:27,853 AND THEN WE HAVE TO THINK HOW DO 2746 01:58:27,853 --> 01:58:29,588 WE TRANSLATE FIGURING OUT HOW 2747 01:58:29,588 --> 01:58:31,556 THAT APPLIES TO THE PATIENTS AS 2748 01:58:31,556 --> 01:58:32,724 PROVIDERS, TO HELP DETERMINE 2749 01:58:32,724 --> 01:58:34,392 WHAT THE BEST COURSE OF ACTION 2750 01:58:34,392 --> 01:58:34,893 IS. 2751 01:58:34,893 --> 01:58:40,766 AND SO THIS KIND OF HAPPENS IN A 2752 01:58:40,766 --> 01:58:43,969 COUPLE WAYS. 2753 01:58:43,969 --> 01:58:46,571 WE HAVE CLINICAL GUIDELINES FROM 2754 01:58:46,571 --> 01:58:48,140 SOCIETIES THAT MANIFEST, SAY AT 2755 01:58:48,140 --> 01:58:50,142 A HOSPITAL OR CLINIC LEVEL. 2756 01:58:50,142 --> 01:58:52,377 IF THOSE GUIDELINES APPLY TO THE 2757 01:58:52,377 --> 01:58:56,515 PATIENT AT HAND WE APPLY THEM. 2758 01:58:56,515 --> 01:58:58,016 IN MANY CASES CONTEXT SUGGESTS 2759 01:58:58,016 --> 01:59:02,020 MAY NOT DIRECTLY APPLY. 2760 01:59:02,020 --> 01:59:03,989 SO THEN DRAW UPON EXPERIENCE AND 2761 01:59:03,989 --> 01:59:05,624 SCIENTIFIC LITERATURE TO FIGURE 2762 01:59:05,624 --> 01:59:07,959 OUT WHAT THAT COURSE OF ACTION. 2763 01:59:07,959 --> 01:59:09,127 THAT'S WHEN YOU TRANSITION FROM 2764 01:59:09,127 --> 01:59:14,299 EVIDENCE-BASED PROTOCOLS TO THIS 2765 01:59:14,299 --> 01:59:16,134 NOTION OF CLINICAL EQUIPOISE. 2766 01:59:16,134 --> 01:59:17,435 IN MANY CASES STRAIGHTFORWARD TO 2767 01:59:17,435 --> 01:59:19,437 FIGURE OUT THE RIGHT COURSE OF 2768 01:59:19,437 --> 01:59:21,173 ACTION BUT WHEN THERE'S A 2769 01:59:21,173 --> 01:59:23,008 BALANCE IN TERMS OF RISK/BENEFIT 2770 01:59:23,008 --> 01:59:28,747 PROFILE YOU'RE IN CLINICAL HE 2771 01:59:28,747 --> 01:59:29,614 EQUIPOISE, GETTING AS MUCH 2772 01:59:29,614 --> 01:59:33,385 INFORMATION ABOUT THE PATIENT AS 2773 01:59:33,385 --> 01:59:34,386 POSSIBLE, PERSONALIZATION 2774 01:59:34,386 --> 01:59:35,987 COMPONENT, AND OBTAINING 2775 01:59:35,987 --> 01:59:36,688 CONSENT, CONVERSATION, 2776 01:59:36,688 --> 01:59:37,622 COMMUNICATING ABOUT RISKS AND 2777 01:59:37,622 --> 01:59:40,325 BENEFITS IS PART OF THIS SECOND 2778 01:59:40,325 --> 01:59:45,397 PART OF CLINICAL DECISION 2779 01:59:45,397 --> 01:59:45,630 MAKING. 2780 01:59:45,630 --> 01:59:46,798 I'D LIKE TO OVERLAY NOTION OF 2781 01:59:46,798 --> 01:59:49,935 RISK ON TOP OF THIS CLINICAL 2782 01:59:49,935 --> 01:59:50,669 DECISION-MAKING PROCESS. 2783 01:59:50,669 --> 01:59:53,605 AND IT KIND OF COVERS ALL 2784 01:59:53,605 --> 01:59:56,341 SORTS -- ALL ASPECTS OF THIS 2785 01:59:56,341 --> 01:59:56,608 DIAGRAM. 2786 01:59:56,608 --> 01:59:58,476 SO INFORMALLY, WE THINK ABOUT 2787 01:59:58,476 --> 02:00:01,213 RISK, WE CAN THINK ABOUT RISK AS 2788 02:00:01,213 --> 02:00:02,781 THE PROBABILITY OF OUTCOMES, 2789 02:00:02,781 --> 02:00:04,616 SEQUELAE OF DISEASE, GIVEN THE 2790 02:00:04,616 --> 02:00:08,687 PATIENT INFORMATION WE HAVE 2791 02:00:08,687 --> 02:00:09,120 AVAILABLE. 2792 02:00:09,120 --> 02:00:11,323 AND THIS MANIFESTS WITH RESPECT 2793 02:00:11,323 --> 02:00:13,425 TO TREATMENT EFFECT, OR RISK 2794 02:00:13,425 --> 02:00:16,261 DIFFERENCES, MIGHT BE LOOKING AT 2795 02:00:16,261 --> 02:00:18,096 THIS QUANTITY UNDER DIFFERENT 2796 02:00:18,096 --> 02:00:19,664 COURSES OF ACTION, POTENTIAL 2797 02:00:19,664 --> 02:00:22,534 INTERVENTIONS, WE CAN DECORATE 2798 02:00:22,534 --> 02:00:24,169 QUANTITY IN THE LANGUAGE OF 2799 02:00:24,169 --> 02:00:27,272 CAUSAL INFERENCE, MIGHT WANT TO 2800 02:00:27,272 --> 02:00:29,708 BUILD THE BEST PREDICTIVE MODELS 2801 02:00:29,708 --> 02:00:31,476 USING LARGE OBSERVATIONAL 2802 02:00:31,476 --> 02:00:34,045 STUDIES OR OTHER INFORMATION TO 2803 02:00:34,045 --> 02:00:34,779 BUILD PREDICTIVE MODELS THAT MAY 2804 02:00:34,779 --> 02:00:36,948 APPLY TO THE PATIENT THAT WE 2805 02:00:36,948 --> 02:00:38,617 HAVE AT HAND AND THEN AGAIN WE 2806 02:00:38,617 --> 02:00:42,988 HAVE THIS EXERCISE, WE HAVE A 2807 02:00:42,988 --> 02:00:44,689 WHOLE SUITE OF RECOMMENDATIONS 2808 02:00:44,689 --> 02:00:46,791 FROM RANDOMIZED TRIALS AS WELL 2809 02:00:46,791 --> 02:00:47,993 AS PREDICTIVE MODELS AND HAVE TO 2810 02:00:47,993 --> 02:00:49,461 FIGURE OUT DO THEY APPLY, 2811 02:00:49,461 --> 02:00:53,598 TRANSPORT TO THE PERSON THAT I'M 2812 02:00:53,598 --> 02:00:56,001 TAKING CARE OF. 2813 02:00:56,001 --> 02:00:58,436 SO, WHEN MY GROUP THINKS ABOUT 2814 02:00:58,436 --> 02:01:00,038 RISK, WE TYPICALLY THINK ABOUT 2815 02:01:00,038 --> 02:01:02,040 IT IN THE TEMPORAL SETTING. 2816 02:01:02,040 --> 02:01:04,442 YET ANOTHER OVERLAY HERE. 2817 02:01:04,442 --> 02:01:05,911 SO TRANSITIONING FROM THE IF 2818 02:01:05,911 --> 02:01:07,245 QUESTION TO THE WHEN QUESTION, 2819 02:01:07,245 --> 02:01:08,914 INSTEAD OF THE PROBABILITY OF 2820 02:01:08,914 --> 02:01:11,683 WHY GIVEN X, WE'RE NOW THINKING 2821 02:01:11,683 --> 02:01:12,918 ABOUT SURVIVAL, THINKING ABOUT 2822 02:01:12,918 --> 02:01:16,087 WHEN DOES THIS EVENT OCCUR, AND 2823 02:01:16,087 --> 02:01:17,923 WHEN WE TRANSITION FROM THE IF 2824 02:01:17,923 --> 02:01:21,059 TO THE WHEN QUESTION THIS 2825 02:01:21,059 --> 02:01:22,127 POTENTIALLY CHANGES BOTH OUR 2826 02:01:22,127 --> 02:01:24,462 FEATURES AND OUR OUTCOMES. 2827 02:01:24,462 --> 02:01:26,331 THE OUTCOME, ONE OF THE 2828 02:01:26,331 --> 02:01:27,632 MOTIVATIONS FOR MOVING INTO THIS 2829 02:01:27,632 --> 02:01:30,735 KIND OF SURVIVAL ANALYSIS SETUP 2830 02:01:30,735 --> 02:01:32,904 IS THAT OFTENTIMES IN LAUNCHING 2831 02:01:32,904 --> 02:01:34,739 STUDIES WE DON'T GET TO SEE WHAT 2832 02:01:34,739 --> 02:01:36,841 HAPPENS TO THE INDIVIDUAL. 2833 02:01:36,841 --> 02:01:41,313 SO WE WANT TO MAKE MODELING 2834 02:01:41,313 --> 02:01:42,213 ASSUMPTION OR SPECIFY A MODEL 2835 02:01:42,213 --> 02:01:45,283 THAT ALLOWS YOU TO CAPTURE THAT. 2836 02:01:45,283 --> 02:01:46,785 LIKEWISE, WHEN WE HAVE 2837 02:01:46,785 --> 02:01:49,120 INFORMATION COMING ACROSS 2838 02:01:49,120 --> 02:01:52,023 TIMING, IN A DYNAMIC WAY, THE 2839 02:01:52,023 --> 02:01:52,624 FUTURE REPRESENTATIONS CHANGE, 2840 02:01:52,624 --> 02:01:55,360 PERHAPS X TURNS INTO AN X STAR, 2841 02:01:55,360 --> 02:01:59,030 X STAR IS SOME OTHER DOMAIN 2842 02:01:59,030 --> 02:02:00,098 BECAUSE INFORMATION IS REVEALED 2843 02:02:00,098 --> 02:02:03,535 ACROSS TIME THAT CHANGES OUR 2844 02:02:03,535 --> 02:02:05,470 BEST REPRESENTATION OF THE 2845 02:02:05,470 --> 02:02:07,205 PATIENT INFORMATION THAT INFORMS 2846 02:02:07,205 --> 02:02:07,572 DECISION MAKING. 2847 02:02:07,572 --> 02:02:10,275 THIS IS CLEAR MODIFICATIONS WHEN 2848 02:02:10,275 --> 02:02:11,876 WE TRANSITION INTO TEMPORAL 2849 02:02:11,876 --> 02:02:12,477 ANALYSIS. 2850 02:02:12,477 --> 02:02:14,612 ALSO IMPLICATIONS WITH RESPECT 2851 02:02:14,612 --> 02:02:16,214 TO OPTIMIZATION FRAMEWORK. 2852 02:02:16,214 --> 02:02:20,719 SO IF WHEN WE'RE THINKING ABOUT 2853 02:02:20,719 --> 02:02:22,620 HAZARD OPTIMIZATION FOR THESE 2854 02:02:22,620 --> 02:02:27,659 FUNCTIONS CAN OFTEN LEAD TO 2855 02:02:27,659 --> 02:02:29,594 OVEREMPHASIS OR UNDEREMPHASIS. 2856 02:02:29,594 --> 02:02:31,963 SO HERE'S AN EXAMPLE OF SOME 2857 02:02:31,963 --> 02:02:33,331 LONGITUDINAL CLINICAL DATA. 2858 02:02:33,331 --> 02:02:35,467 WHAT I'M SHOWING HERE IS ONE 2859 02:02:35,467 --> 02:02:37,369 PATIENT OVER THE COURSE OF 2860 02:02:37,369 --> 02:02:47,245 ROUGHLY A WEEK IN A HOSPITAL, 2861 02:02:47,245 --> 02:02:48,847 AND DE-IDENTIFY. 2862 02:02:48,847 --> 02:02:50,248 EVERY DOT REPRESENTS THE 2863 02:02:50,248 --> 02:02:52,283 COLLECTION OF SOME MEASUREMENT, 2864 02:02:52,283 --> 02:02:55,687 LIKE A BLOOD PRESSURE, BLOOD 2865 02:02:55,687 --> 02:02:55,920 GLUCOSE. 2866 02:02:55,920 --> 02:02:59,290 SO THIS IS WHAT WE TYPICALLY SEE 2867 02:02:59,290 --> 02:03:00,091 IN ELECTRONIC HEALTH RECORDS 2868 02:03:00,091 --> 02:03:02,927 DATA, PRETTY MUCH A PERVASIVE 2869 02:03:02,927 --> 02:03:05,030 SOURCE OF LONGITUDINAL DATA 2870 02:03:05,030 --> 02:03:07,365 ABOUT PATIENTS AT THIS POINT. 2871 02:03:07,365 --> 02:03:08,666 YOU KNOW, WITH INCREASE IN USE 2872 02:03:08,666 --> 02:03:10,702 OVER THE LAST 20 YEARS. 2873 02:03:10,702 --> 02:03:12,937 AND WE TYPICALLY GET THIS IN A 2874 02:03:12,937 --> 02:03:14,439 STRUCTURED FORMAT DRAWING FROM 2875 02:03:14,439 --> 02:03:16,875 TABLES SUCH AS ADMISSION, 2876 02:03:16,875 --> 02:03:17,475 CHARTING, DIAGNOSES, LABS, 2877 02:03:17,475 --> 02:03:19,277 MEDICATIONS, SO FORTH. 2878 02:03:19,277 --> 02:03:20,345 THIS NICELY REPRESENTS REAL 2879 02:03:20,345 --> 02:03:23,882 WORLD DATA BECAUSE IT'S 2880 02:03:23,882 --> 02:03:24,849 COLLECTED AS NEEDED, REFLECTED 2881 02:03:24,849 --> 02:03:27,152 AT LARGE SCALE OF REAL WORLD 2882 02:03:27,152 --> 02:03:28,586 SITUATIONS AND POPULATIONS, BUT 2883 02:03:28,586 --> 02:03:32,757 HAS A NUMBER OF REALLY 2884 02:03:32,757 --> 02:03:33,291 STATISTICAL CHALLENGES AND 2885 02:03:33,291 --> 02:03:34,459 INCOMPLETE, INCOMPLETE IN A WAY 2886 02:03:34,459 --> 02:03:36,561 THAT'S NOT AT RANDOM AS IN THE 2887 02:03:36,561 --> 02:03:37,962 MODEL DOESN'T KNOW WHY SOME 2888 02:03:37,962 --> 02:03:39,097 THINGS ARE COLLECTED OR NOT, NOT 2889 02:03:39,097 --> 02:03:40,965 COLLECTED IN A REGULAR FASHION, 2890 02:03:40,965 --> 02:03:43,268 SO WE DON'T HAVE THE INFORMATION 2891 02:03:43,268 --> 02:03:46,071 NECESSARY TO UNDERSTAND THOSE 2892 02:03:46,071 --> 02:03:46,337 PROCESSES. 2893 02:03:46,337 --> 02:03:47,806 WE'RE JUST GIVEN CAUTIONS, 2894 02:03:47,806 --> 02:03:48,506 MEASUREMENTS WHEN THEY ACTUALLY 2895 02:03:48,506 --> 02:03:49,808 DO HAPPEN. 2896 02:03:49,808 --> 02:03:52,644 SO THIS CAN AFFECT THE MODELING 2897 02:03:52,644 --> 02:03:56,915 THAT TRANSPORTABILITY OF THE 2898 02:03:56,915 --> 02:03:59,217 MODEL THAT WE BUILD ON THESE 2899 02:03:59,217 --> 02:03:59,784 DATA. 2900 02:03:59,784 --> 02:04:01,619 IF YOU HAVE DIFFERENT 2901 02:04:01,619 --> 02:04:02,487 COLLECTIONS AT ANOTHER LOCATION 2902 02:04:02,487 --> 02:04:05,223 MODEL THAT YOU PREDICT AT YOUR 2903 02:04:05,223 --> 02:04:07,525 ORIGIN SITE MIGHT NOT TRANSFER 2904 02:04:07,525 --> 02:04:10,695 TO THE DESTINATION BECAUSE OF 2905 02:04:10,695 --> 02:04:10,929 THIS. 2906 02:04:10,929 --> 02:04:13,765 IN THE CONTEXT OF ALL OF THIS MY 2907 02:04:13,765 --> 02:04:16,501 GROUP FOCUSES ON THREE AREAS. 2908 02:04:16,501 --> 02:04:20,572 I'LL HIGHLIGHT THE FIRST TWO AT 2909 02:04:20,572 --> 02:04:22,474 A HIGH LEVEL, JUST A TASTE, AND 2910 02:04:22,474 --> 02:04:30,315 DO A DEEPER DIVE INTO THE 2911 02:04:30,315 --> 02:04:32,717 TEXTUAL COMPONENT. 2912 02:04:32,717 --> 02:04:32,917 OKAY. 2913 02:04:32,917 --> 02:04:35,820 SO, IN THE FIRST EXAMPLE, WE 2914 02:04:35,820 --> 02:04:39,023 WERE INTERESTED, AS PART OF THE 2915 02:04:39,023 --> 02:04:43,728 STUDY, INVOLVED IN JUST 2916 02:04:43,728 --> 02:04:44,996 IDENTIFYING SUBGROUPS OF SEPSIS, 2917 02:04:44,996 --> 02:04:46,264 THIS IS JAMA, 2019. 2918 02:04:46,264 --> 02:04:50,335 THE IDEA HERE IS THAT SEPSIS, 2919 02:04:50,335 --> 02:04:52,303 WHICH IS A BODY'S RESPONSE TO 2920 02:04:52,303 --> 02:04:54,172 INFECTION, THAT CAN LEAD TO 2921 02:04:54,172 --> 02:04:57,342 ORGAN FAILURE AND DEATH, IS THAT 2922 02:04:57,342 --> 02:04:59,277 SEPSIS IS A NOTORIOUSLY 2923 02:04:59,277 --> 02:05:00,612 HETEROGENEOUS DISEASE. 2924 02:05:00,612 --> 02:05:04,315 THE SITE OF INFECTION, THE 2925 02:05:04,315 --> 02:05:07,018 COMORBIDITY PROFILE, ORGANS AT 2926 02:05:07,018 --> 02:05:09,120 RISK MAY ALL DIFFER, AND SO WHAT 2927 02:05:09,120 --> 02:05:11,689 WE HAD SEEN IN THE LITERATURE IS 2928 02:05:11,689 --> 02:05:14,792 THAT, YOU KNOW, TWO DECADES 2929 02:05:14,792 --> 02:05:16,628 WORTH OF RANDOMIZED CONTROLLED 2930 02:05:16,628 --> 02:05:18,897 TRIALS THAT FAILED TO FIND A 2931 02:05:18,897 --> 02:05:19,998 POSITIVE FINDING FOR NEW 2932 02:05:19,998 --> 02:05:21,499 TREATMENT OR THERAPEUTIC OR 2933 02:05:21,499 --> 02:05:22,167 FAILED TO REPLICATE IN 2934 02:05:22,167 --> 02:05:23,368 SUBSEQUENT TRIALS. 2935 02:05:23,368 --> 02:05:27,238 SO THE HYPOTHESIS IS THAT IF YOU 2936 02:05:27,238 --> 02:05:28,740 CAN IDENTIFY SUBGROUPS THAT ARE 2937 02:05:28,740 --> 02:05:30,141 HOMOGENOUS IN NATURE YOU MIGHT 2938 02:05:30,141 --> 02:05:33,111 BE ABLE TO USE TARGETED 2939 02:05:33,111 --> 02:05:35,113 THERAPEUTICS THAT WOULD SUCCEED 2940 02:05:35,113 --> 02:05:37,215 WITHIN THAT SUBGROUP. 2941 02:05:37,215 --> 02:05:38,750 THIS PARTICULAR WORK IDENTIFIED 2942 02:05:38,750 --> 02:05:40,251 FOUR SUBGROUPS BASED ON 2943 02:05:40,251 --> 02:05:42,253 INFORMATION COLLECTED WITHIN THE 2944 02:05:42,253 --> 02:05:44,355 FIRST SIX HOURS OF ADMISSIONS. 2945 02:05:44,355 --> 02:05:46,624 NOW, ONE OF THE LIMITATIONS OF 2946 02:05:46,624 --> 02:05:50,261 THIS STUDY IS THAT THE ONLY TIME 2947 02:05:50,261 --> 02:05:51,095 THIS CLUSTERING HAPPENED, OR 2948 02:05:51,095 --> 02:05:53,198 DATA THAT WAS USED, WAS FROM THE 2949 02:05:53,198 --> 02:05:54,966 FIRST SIX HOURS ONLY. 2950 02:05:54,966 --> 02:05:56,834 AND SO IT'S QUITE POSSIBLE THAT 2951 02:05:56,834 --> 02:05:59,337 A PERSON WHO INITIALLY WAS IN 2952 02:05:59,337 --> 02:06:00,972 LET'S SAY SUBGROUP ALPHA 2953 02:06:00,972 --> 02:06:05,310 ACTUALLY IN A COUPLE HOURS DUE 2954 02:06:05,310 --> 02:06:07,078 TO DETERIORATION WOULD THEN BE 2955 02:06:07,078 --> 02:06:08,213 PLACED INTO ANOTHER GROUP AT 2956 02:06:08,213 --> 02:06:09,547 HIGHER RISK, LET'S SAY GROUP 2957 02:06:09,547 --> 02:06:09,781 DELTA. 2958 02:06:09,781 --> 02:06:13,117 BECAUSE WE WANT TO BE ABLE TO 2959 02:06:13,117 --> 02:06:15,820 HAVE A PERSON, YOU KNOW, NOT DO 2960 02:06:15,820 --> 02:06:17,822 THIS HAVE THIS JUMPING AS WE SEE 2961 02:06:17,822 --> 02:06:21,059 IT, WE WANTED TO BE ABLE TO 2962 02:06:21,059 --> 02:06:22,894 CLUSTER ON SNAPSHOTS. 2963 02:06:22,894 --> 02:06:26,097 SO FOLLOW-UP WORK ALLOWED US TO 2964 02:06:26,097 --> 02:06:28,132 DO THIS. 2965 02:06:28,132 --> 02:06:30,168 WE INTRODUCED A TEMPORAL 2966 02:06:30,168 --> 02:06:30,902 SUPERVISED CONTRASTED LEARNING 2967 02:06:30,902 --> 02:06:33,504 METHOD INSTEAD OF TAKING THE 2968 02:06:33,504 --> 02:06:35,440 INFORMATION JUST AT THE TIME, WE 2969 02:06:35,440 --> 02:06:38,977 TOOK FOUR-HOUR SNAPSHOTS AND 2970 02:06:38,977 --> 02:06:40,612 USED THAT INFORMATION, PLACED 2971 02:06:40,612 --> 02:06:43,248 INTO EMBEDDING SPACE. 2972 02:06:43,248 --> 02:06:44,849 THIS SPACE WAS CHARACTERIZED BY 2973 02:06:44,849 --> 02:06:47,085 HAVING EACH OF THE SNAPSHOTS OF 2974 02:06:47,085 --> 02:06:48,253 THE PATIENT BEING LOCALIZED 2975 02:06:48,253 --> 02:06:52,457 CLOSE TO ONE ANOTHER AND THEN 2976 02:06:52,457 --> 02:06:53,358 EMBEDDING SHOULD HAVE SIMILAR 2977 02:06:53,358 --> 02:06:55,693 REPRESENTATION WITH RESPECT TO 2978 02:06:55,693 --> 02:06:56,694 FEATURES, ALSO WITH RESPECT TO 2979 02:06:56,694 --> 02:06:57,161 RISK. 2980 02:06:57,161 --> 02:07:00,898 AS A RESULT WE WERE ABLE TO 2981 02:07:00,898 --> 02:07:02,800 IDENTIFY TWO SUBGROUPS THAT WERE 2982 02:07:02,800 --> 02:07:05,470 HOMOGENOUS WITH RESPECT TO RISK, 2983 02:07:05,470 --> 02:07:07,338 BUT HAD VARYING PROFILES. 2984 02:07:07,338 --> 02:07:10,575 SO WHAT I'M SHOWING IN THE HEAT 2985 02:07:10,575 --> 02:07:11,276 MAP IS PROBABILITY OF 2986 02:07:11,276 --> 02:07:12,610 REMEMBERING WHAT'S IN THE 2987 02:07:12,610 --> 02:07:15,613 SUBGROUP TO HAVE THAT PARTICULAR 2988 02:07:15,613 --> 02:07:17,148 LABORATORY CHARACTERISTICS AND 2989 02:07:17,148 --> 02:07:18,249 DESPITE PEOPLE HAVING VARIOUS 2990 02:07:18,249 --> 02:07:21,519 SIMILAR RISK WITH RESPECT TO 2991 02:07:21,519 --> 02:07:23,187 90-DAY MORTALITY, THEY ACTUALLY 2992 02:07:23,187 --> 02:07:27,492 LOOKED VERY DIFFERENT WITH 2993 02:07:27,492 --> 02:07:30,295 RESPECT TO LABORATORY CHARACTER. 2994 02:07:30,295 --> 02:07:35,400 ANOTHER WORK OF OURS HAS FOCUSED 2995 02:07:35,400 --> 02:07:36,601 ON CHARACTERIZING THE HAZARD 2996 02:07:36,601 --> 02:07:38,336 FUNCTION AND OPTIMIZATIONS 2997 02:07:38,336 --> 02:07:39,904 AROUND THE LIKELIHOOD OF THAT 2998 02:07:39,904 --> 02:07:41,973 HAZARD, AND MOTIVATION HERE IS 2999 02:07:41,973 --> 02:07:46,311 THAT PEOPLE WHO ARE LOW RISK, 3000 02:07:46,311 --> 02:07:48,546 OFTEN A PLURALITY OF PATIENTS 3001 02:07:48,546 --> 02:07:51,282 ARE NOT WELL REPRESENTED WHEN 3002 02:07:51,282 --> 02:07:54,152 USING OBJECTIVE FUNCTIONS, THAT 3003 02:07:54,152 --> 02:07:55,820 IS OBJECTIVE FUNCTIONS WILL 3004 02:07:55,820 --> 02:07:59,257 FOCUS ON HIGHEST RISK 3005 02:07:59,257 --> 02:07:59,557 INDIVIDUALS. 3006 02:07:59,557 --> 02:08:05,863 IF YOU THESE TO MAKE 3007 02:08:05,863 --> 02:08:06,764 RECOMMENDATIONS FOR PLURALITY, 3008 02:08:06,764 --> 02:08:09,167 MAY NOT BE OPTIMIZED FOR MAKING 3009 02:08:09,167 --> 02:08:11,636 RECOMMENDATIONS, THE WORK HELPED 3010 02:08:11,636 --> 02:08:12,837 HIGHLIGHT THAT ASPECT. 3011 02:08:12,837 --> 02:08:15,506 AND THEN IF YOU WANT TO NOT ONLY 3012 02:08:15,506 --> 02:08:17,775 PREDICT WELL, BUT YOU ALSO WANT 3013 02:08:17,775 --> 02:08:19,644 TO HAVE SOME OTHER 3014 02:08:19,644 --> 02:08:21,846 CHARACTERISTIC PROPERTIES, WE'VE 3015 02:08:21,846 --> 02:08:26,284 INTRODUCED A SERIES OF WORKS 3016 02:08:26,284 --> 02:08:30,154 FOCUSED ON HAVING HIGH CAPACITY 3017 02:08:30,154 --> 02:08:31,322 PERFORMANCE, NOT TOO DIFFERENT 3018 02:08:31,322 --> 02:08:32,824 ACROSS SUBGROUPS OF INTEREST. 3019 02:08:32,824 --> 02:08:36,527 IN PARTICULAR, INSTEAD OF USING 3020 02:08:36,527 --> 02:08:39,230 COMMONLY USED NOTIONS, LIKE E. 3021 02:08:39,230 --> 02:08:40,098 COLI'S ODDS OF STATISTICAL 3022 02:08:40,098 --> 02:08:42,133 PARITY, THESE TIMES OF 3023 02:08:42,133 --> 02:08:43,968 DEFINITIONS ACTUALLY DON'T WORK 3024 02:08:43,968 --> 02:08:45,603 IN THE RIGHT SENSORSHIP SETTING 3025 02:08:45,603 --> 02:08:46,904 WHERE OUTCOME IS UNKNOWN. 3026 02:08:46,904 --> 02:08:54,879 WE DON'T KNOW WHAT HAPPENS TO 3027 02:08:54,879 --> 02:08:55,179 INDIVIDUALS. 3028 02:08:55,179 --> 02:09:02,153 WE'VE IDENTIFIED TERMINOLOGY TO 3029 02:09:02,153 --> 02:09:04,455 HELP YOU IDENTIFY, CONCORDANCE, 3030 02:09:04,455 --> 02:09:06,190 PARITY, MODELS TO PREDICT BUT 3031 02:09:06,190 --> 02:09:14,031 HAVE LOW DIFFERENCES ACROSS 3032 02:09:14,031 --> 02:09:14,298 SUBGROUPS. 3033 02:09:14,298 --> 02:09:14,499 OKAY. 3034 02:09:14,499 --> 02:09:17,902 I'M GOING TO DIVE INTO THE 3035 02:09:17,902 --> 02:09:18,936 TEXTUAL LONGITUDINAL DATA 3036 02:09:18,936 --> 02:09:20,772 COMPONENT FIRST BUT I WANTED TO 3037 02:09:20,772 --> 02:09:24,075 HIGHLIGHT THAT WE JUST RECENTLY 3038 02:09:24,075 --> 02:09:29,547 HAD A PAPER ACCEPTED, AND THE MY 3039 02:09:29,547 --> 02:09:33,351 Ph.D. STUDENT CO-ADVISED 3040 02:09:33,351 --> 02:09:35,353 PRESENTING HIS POSTER ON 3041 02:09:35,353 --> 02:09:36,087 ANTIPSYCHOTIC. 3042 02:09:36,087 --> 02:09:37,722 SO, NOW WE'RE TALKING ABOUT 3043 02:09:37,722 --> 02:09:40,291 TEXTUAL TIME PERIOD, I WANTED TO 3044 02:09:40,291 --> 02:09:44,262 GIVE YOU A DESCRIPTION OF THIS 3045 02:09:44,262 --> 02:09:44,462 TASK. 3046 02:09:44,462 --> 02:09:47,465 SO, WE HAVE MANY CASE REPORTS, 3047 02:09:47,465 --> 02:09:48,166 DISCHARGE SUMMARIES REPRESENTED 3048 02:09:48,166 --> 02:09:50,435 BY THE BLURB ON THE LEFT. 3049 02:09:50,435 --> 02:09:52,470 AND THESE ARE DOCUMENTS THAT ARE 3050 02:09:52,470 --> 02:09:58,209 TIME STAMPED AT THE END OF THE 3051 02:09:58,209 --> 02:10:01,245 ATTEMPT, AFTER THE FACT. 3052 02:10:01,245 --> 02:10:03,047 HOWEVER, THERE ARE MANY MENTIONS 3053 02:10:03,047 --> 02:10:04,015 OF -- WITHIN THE CASE REPORT 3054 02:10:04,015 --> 02:10:07,251 THAT CAN BE TIME STAMPED WITH 3055 02:10:07,251 --> 02:10:12,123 MUCH GREATER GRANULARITY. 3056 02:10:12,123 --> 02:10:13,224 SO, THE TASK IS TO TAKE MENTIONS 3057 02:10:13,224 --> 02:10:16,561 AND PLACE THEM ON A TIMELINE 3058 02:10:16,561 --> 02:10:18,429 BECAUSE THIS INFORMATION WOULD 3059 02:10:18,429 --> 02:10:20,631 BE ABLE TO PROVIDERS TRYING TO 3060 02:10:20,631 --> 02:10:21,532 DO RISK ASSESSMENTS AT THE POINT 3061 02:10:21,532 --> 02:10:22,400 OF CARE. 3062 02:10:22,400 --> 02:10:23,968 HOWEVER, THEY DON'T END UP 3063 02:10:23,968 --> 02:10:25,736 MAKING IT INTO THE LONGITUDINAL 3064 02:10:25,736 --> 02:10:27,538 RISK MODELS BECAUSE THEY HAVE 3065 02:10:27,538 --> 02:10:29,073 BEEN TIME STAMPED AFTER THE 3066 02:10:29,073 --> 02:10:29,273 FACT. 3067 02:10:29,273 --> 02:10:32,009 SO OUR GOAL IS TO BE ABLE TO 3068 02:10:32,009 --> 02:10:32,543 EXTRACT THIS INFORMATION, 3069 02:10:32,543 --> 02:10:35,246 PROVIDED AS A TIMELINE SO WE CAN 3070 02:10:35,246 --> 02:10:40,718 BUILD MODELS ON THE TEXTUAL TIME 3071 02:10:40,718 --> 02:10:41,786 PERIOD, ALONGSIDE TABULAR 3072 02:10:41,786 --> 02:10:42,520 PRESENTATION. 3073 02:10:42,520 --> 02:10:44,755 SO, THIS IS A MOTIVATING SLIDE 3074 02:10:44,755 --> 02:10:46,290 ABOUT WHY THIS IS IMPORTANT. 3075 02:10:46,290 --> 02:10:50,094 THIS PAPER WHICH WAS PUBLISHED 3076 02:10:50,094 --> 02:10:55,766 EARLIER THIS YEAR IN JMIR 3077 02:10:55,766 --> 02:10:59,270 BASICALLY MAPS TEXTUAL CASE 3078 02:10:59,270 --> 02:11:01,973 REPORTS TO CLINICAL ENTITIES, 3079 02:11:01,973 --> 02:11:08,246 FROM UML ONTOLOGY, AND THEN THE 3080 02:11:08,246 --> 02:11:17,522 SAME THING FOR TABULATING. 3081 02:11:17,522 --> 02:11:20,958 ONLY 13 WERE PRESENT, AND THIS 3082 02:11:20,958 --> 02:11:24,896 GIVES A SINCE THE TWO DATA FORMS 3083 02:11:24,896 --> 02:11:26,664 ARE COMPLEMENTARY YET RESEARCH 3084 02:11:26,664 --> 02:11:28,099 COMMITTEE FOCUSES ON HEAVILY ON 3085 02:11:28,099 --> 02:11:33,371 THE TABULAR DATA ALONE BECAUSE 3086 02:11:33,371 --> 02:11:34,038 OF ITS AVAILABILITY. 3087 02:11:34,038 --> 02:11:36,107 IT'S BEEN OUR GOAL TO HAVE 3088 02:11:36,107 --> 02:11:39,644 MODELS BUILT ON THE BEST BUILT 3089 02:11:39,644 --> 02:11:41,012 DATA, RICHEST FROM PREVIOUS 3090 02:11:41,012 --> 02:11:43,714 SLIDES, WE WANT TO PROVIDE THIS 3091 02:11:43,714 --> 02:11:48,586 NEW DATA SOURCE TO IMPROVE THE 3092 02:11:48,586 --> 02:11:50,755 PREDICTED MODELS. 3093 02:11:50,755 --> 02:11:52,456 HERE IS THE DEFINITION, TEXTUAL 3094 02:11:52,456 --> 02:11:57,194 TIME SERIES IS A SEQUENCE OF 3095 02:11:57,194 --> 02:11:59,797 EVENTS, HERE WE'RE DEFINING 3096 02:11:59,797 --> 02:12:04,302 EVENT AS CONTINUOUS SPAN, TIME 3097 02:12:04,302 --> 02:12:06,437 LOCALIZABLE EVENT, AND THE TIME, 3098 02:12:06,437 --> 02:12:08,205 EVENT TIME, RELATIVE TO INITIAL 3099 02:12:08,205 --> 02:12:09,073 PRESENTATION. 3100 02:12:09,073 --> 02:12:11,742 FOR THE SAKE OF TIME, I'M GOING 3101 02:12:11,742 --> 02:12:15,780 TO FOCUS ON THE CONTRAST HERE. 3102 02:12:15,780 --> 02:12:18,149 EVENTS AND TIMES WE'RE USING ARE 3103 02:12:18,149 --> 02:12:19,483 SLIGHTLY DIFFERENT FROM THOSE 3104 02:12:19,483 --> 02:12:22,453 THAT ARE USED, AND THE CLINICAL 3105 02:12:22,453 --> 02:12:23,187 TEMPORAL COMMUNITY. 3106 02:12:23,187 --> 02:12:25,957 THERE'S BEEN A LOT OF WORK IN 3107 02:12:25,957 --> 02:12:28,326 ANY COMMUNITY ON TEMPORAL 3108 02:12:28,326 --> 02:12:29,460 RELATIONS, CLINICAL CONCEPTS. 3109 02:12:29,460 --> 02:12:31,996 AND HERE BECAUSE WE WANT TO DO 3110 02:12:31,996 --> 02:12:35,132 THESE DOWNSTREAM TASKS OF 3111 02:12:35,132 --> 02:12:35,800 FORECASTING CAUSAL INFERENCE WE 3112 02:12:35,800 --> 02:12:38,102 DON'T WANT TO DO NECESSARILY A 3113 02:12:38,102 --> 02:12:41,505 MAPPING TO ONTOLOGY. 3114 02:12:41,505 --> 02:12:43,040 WE WANT TO RETAIN CONTEXTUAL 3115 02:12:43,040 --> 02:12:44,442 INFORMATION LOCALIZED AT THIS 3116 02:12:44,442 --> 02:12:46,310 TIME POINT SO OTHER TYPES OF 3117 02:12:46,310 --> 02:12:55,252 MODELS CAN PERFORM AS BEST AS 3118 02:12:55,252 --> 02:12:55,519 POSSIBLE. 3119 02:12:55,519 --> 02:12:59,190 SIMILARLY, THIS NLT COMMUNITYE 3120 02:12:59,190 --> 02:13:01,125 TEMPORAL RELATION, ASKS WHETHER 3121 02:13:01,125 --> 02:13:03,260 EVENT A PRECEDES EVENT B, YES OR 3122 02:13:03,260 --> 02:13:07,565 NO, AND HERE WE WANT TO CONDUCT 3123 02:13:07,565 --> 02:13:08,199 SURVIVAL ANALYSIS, LONGITUDINAL 3124 02:13:08,199 --> 02:13:12,937 ANALYSIS, WE WANT THE TIMING 3125 02:13:12,937 --> 02:13:16,240 RELATIVE TO THIS. 3126 02:13:16,240 --> 02:13:18,342 SO, WE'RE GOING TO USE LARGE 3127 02:13:18,342 --> 02:13:19,610 LANGUAGE MODELS TO GET 3128 02:13:19,610 --> 02:13:21,712 ANNOTATIONS AT SCALE BUT NEED 3129 02:13:21,712 --> 02:13:22,913 SOME GROUND TRUTH. 3130 02:13:22,913 --> 02:13:26,517 FIRST TO DO THIS WE CONDUCTED 3131 02:13:26,517 --> 02:13:31,722 SEVERAL PILOTS TO COLLECT 3132 02:13:31,722 --> 02:13:32,590 INFORMATION FROM CLINICAL 3133 02:13:32,590 --> 02:13:34,925 ANNOTATORS, MYSELF AND OTHER 3134 02:13:34,925 --> 02:13:36,060 CLINICAL PROVIDERS. 3135 02:13:36,060 --> 02:13:37,962 AND OUR JOB ESSENTIALLY WAS TO 3136 02:13:37,962 --> 02:13:40,064 IDENTIFY EVENTS FROM THE TEXT 3137 02:13:40,064 --> 02:13:42,400 AND THEN SELECT TIME INTERVALS 3138 02:13:42,400 --> 02:13:43,968 BASED ON TABULAR REPRESENTATIONS 3139 02:13:43,968 --> 02:13:47,471 OR JUST TIMELINES AND THEN 3140 02:13:47,471 --> 02:13:50,107 REPEATEDLY SELECT ANNOTATES TO 3141 02:13:50,107 --> 02:13:52,143 GENERATE THIS CORPUS OF DATA. 3142 02:13:52,143 --> 02:13:54,045 THEN WE'RE ABLE TO COMPARE THESE 3143 02:13:54,045 --> 02:13:55,146 MANUAL ANNOTATIONS WHICH ARE 3144 02:13:55,146 --> 02:13:57,014 GOING TO BE SMALL SCALE 3145 02:13:57,014 --> 02:13:58,916 EXPENSIVE TO COLLECT, WITH THE 3146 02:13:58,916 --> 02:13:59,583 ANNOTATIONS WE CAN GET FROM 3147 02:13:59,583 --> 02:14:01,052 LARGE LANGUAGE MODELS. 3148 02:14:01,052 --> 02:14:04,488 SO THIS IS ONE TYPE PLAN THAT 3149 02:14:04,488 --> 02:14:07,525 WE'VE DEVELOPED WHICH HIGH LEVEL 3150 02:14:07,525 --> 02:14:08,259 INVOLVES EXTRACTION, ANNOTATION, 3151 02:14:08,259 --> 02:14:13,030 AND THIS IS FOR THE PubMed 3152 02:14:13,030 --> 02:14:16,467 OPEN ACCESS CORPUS. 3153 02:14:16,467 --> 02:14:18,069 THIS CORPUS HAS 1.5 MILLION 3154 02:14:18,069 --> 02:14:20,938 PUBLISHED MANUSCRIPTS, OF THESE 3155 02:14:20,938 --> 02:14:21,906 WE CAN IDENTIFY APPROXIMATELY 3156 02:14:21,906 --> 02:14:24,375 125,000 CASE REPORTS OF WHICH 3157 02:14:24,375 --> 02:14:26,277 WE'VE ANNOTATED, YOU KNOW, ALL 3158 02:14:26,277 --> 02:14:27,745 OF THEM WITH THE LARGE LANGUAGE 3159 02:14:27,745 --> 02:14:32,616 MODELS TO CONSTRUCT OUR TEXTUAL 3160 02:14:32,616 --> 02:14:37,455 TIME SERIES. 3161 02:14:37,455 --> 02:14:41,492 WE'RE ABLE TO COMPARE AGAINST 3162 02:14:41,492 --> 02:14:43,794 OUR MANUAL SUBSET AND THEN ALSO 3163 02:14:43,794 --> 02:14:50,034 PROVIDE THE ENTIRE CORPUS TO THE 3164 02:14:50,034 --> 02:14:53,504 COMMUNITY, TO APPEND ADDITIONAL 3165 02:14:53,504 --> 02:14:54,171 INFORMATION, OUTCOME 3166 02:14:54,171 --> 02:14:55,172 INFORMATION, ALONGSIDE 3167 02:14:55,172 --> 02:14:55,906 CHARACTERIZATIONS THAT ALLOW 3168 02:14:55,906 --> 02:14:59,343 PEOPLE TO DRILL IN TO IDENTIFY 3169 02:14:59,343 --> 02:14:59,777 COHORTS. 3170 02:14:59,777 --> 02:15:03,080 AND THEN BECAUSE WE'RE DOING 3171 02:15:03,080 --> 02:15:04,682 THIS NEW PROCESS OF, YOU KNOW, 3172 02:15:04,682 --> 02:15:06,450 EXTRACTING EVENTS IN THE WAY WE 3173 02:15:06,450 --> 02:15:08,285 THINK IS RIGHT TO SPECIFY 3174 02:15:08,285 --> 02:15:10,521 ALONGSIDE THE TIMING WE HAVE TO 3175 02:15:10,521 --> 02:15:12,022 DEVELOP OUR OWN ASSESSMENT TOOLS 3176 02:15:12,022 --> 02:15:15,526 TO ASSESS THE QUALITY OF THE 3177 02:15:15,526 --> 02:15:16,761 ANNOTATIONS. 3178 02:15:16,761 --> 02:15:19,463 SO I'LL GIVE THE HIGH LEVEL 3179 02:15:19,463 --> 02:15:20,064 PICTURE OF THIS. 3180 02:15:20,064 --> 02:15:21,999 BASICALLY THERE ARE A COUPLE 3181 02:15:21,999 --> 02:15:27,805 STEPS THAT WE DO, WE TAKE. 3182 02:15:27,805 --> 02:15:29,540 FIRST IS IDENTIFYING TEXT 3183 02:15:29,540 --> 02:15:30,007 MATCHING. 3184 02:15:30,007 --> 02:15:31,809 SO YOU COULD THINK ABOUT 3185 02:15:31,809 --> 02:15:33,711 CHARACTER LEVEL SIMILARITIES BUT 3186 02:15:33,711 --> 02:15:38,215 WE FIND THAT USING SENTENCE 3187 02:15:38,215 --> 02:15:42,253 TRANSFORMING EMBEDDINGS AND 3188 02:15:42,253 --> 02:15:43,587 DISTANCES BETWEEN, SIMILARITY 3189 02:15:43,587 --> 02:15:44,522 WORKS BETTER. 3190 02:15:44,522 --> 02:15:47,758 WE TRIED A BUNCH OF MODELS, TO 3191 02:15:47,758 --> 02:15:50,528 ANNOTATE AND COMPARE THE 3192 02:15:50,528 --> 02:15:53,297 SIMILARITIES, OR WHETHER OR NOT 3193 02:15:53,297 --> 02:15:55,599 THOSE EVENTS MATCH OUR MANUAL 3194 02:15:55,599 --> 02:15:58,669 ANNOTATED EVENTS, AND WHAT WE 3195 02:15:58,669 --> 02:16:06,577 FIND IS THAT USING COSINE 3196 02:16:06,577 --> 02:16:09,580 DISTANCE, 75% MATCH WITH THE 3197 02:16:09,580 --> 02:16:10,815 ONES WE MANUALLY ENTERED. 3198 02:16:10,815 --> 02:16:13,150 HOW DO WE CHOOSE THAT DISTANCE 3199 02:16:13,150 --> 02:16:13,984 LEVEL? 3200 02:16:13,984 --> 02:16:17,354 BASED ON MANUAL REVIEW, WHERE WE 3201 02:16:17,354 --> 02:16:19,824 CAN SEE IN FACT IT CORRESPONDS 3202 02:16:19,824 --> 02:16:21,892 TO THE SAME NOTION. 3203 02:16:21,892 --> 02:16:23,794 ONCE WE'VE LOOKED AT EVENT 3204 02:16:23,794 --> 02:16:26,330 MATCHES WE CAN CONSIDER WHETHER 3205 02:16:26,330 --> 02:16:30,568 OR NOT THE TIMES PRODUCED BY 3206 02:16:30,568 --> 02:16:31,335 THESE LARGE LANGUAGE MODELS 3207 02:16:31,335 --> 02:16:32,736 CORRESPOND TO THE TIME WE 3208 02:16:32,736 --> 02:16:34,271 IDENTIFY IN OUR REPORTS. 3209 02:16:34,271 --> 02:16:39,043 SINCE WE'RE LOOKING AT TIME 3210 02:16:39,043 --> 02:16:40,411 DIFFERENCE, MEAN ABSOLUTE AREA, 3211 02:16:40,411 --> 02:16:41,512 THESE DON'T WORK VERY WELL WHEN 3212 02:16:41,512 --> 02:16:43,113 YOU THINK ABOUT THE DIFFERENT 3213 02:16:43,113 --> 02:16:44,915 TIME SCALES THAT WE'RE WORKING 3214 02:16:44,915 --> 02:16:47,785 OVER, POTENTIALLY YEARS TO 3215 02:16:47,785 --> 02:16:48,118 MINUTES. 3216 02:16:48,118 --> 02:16:49,587 IF YOU THINK ABOUT FURTHER 3217 02:16:49,587 --> 02:16:51,455 ERRORS AT THE LEVEL OF YEARS, 3218 02:16:51,455 --> 02:16:52,790 WE'RE GOING TO DOMINATE THE 3219 02:16:52,790 --> 02:16:55,259 ERROR THAT I MADE AT THE LEVEL 3220 02:16:55,259 --> 02:16:55,793 OF MINUTES. 3221 02:16:55,793 --> 02:16:59,163 WE CAN EITHER DRAW ON, YOU KNOW, 3222 02:16:59,163 --> 02:17:00,798 IDEAS FROM ROBUST STATISTICS 3223 02:17:00,798 --> 02:17:03,901 LIKE MEDIAN ABSOLUTE ERROR OR 3224 02:17:03,901 --> 02:17:05,402 DEVELOP OUR OWN METRIC OF AREA 3225 02:17:05,402 --> 02:17:07,504 UNDER THE LOG TIME CURVE, WHAT 3226 02:17:07,504 --> 02:17:11,542 I'M SHOWING IN THE GRAPH IS THE 3227 02:17:11,542 --> 02:17:14,845 LOG TIME PLUS OF TIME 3228 02:17:14,845 --> 02:17:15,946 DISCREPANCIES BETWEEN LARGE 3229 02:17:15,946 --> 02:17:18,816 LANGUAGE MODEL TIME AND MANUAL 3230 02:17:18,816 --> 02:17:20,417 ANNOTATION OUTCOME. 3231 02:17:20,417 --> 02:17:21,919 WE CAN SEE THAT APPROXIMATELY 3232 02:17:21,919 --> 02:17:25,890 50% OF THESE EVENTS GETTING THE 3233 02:17:25,890 --> 02:17:27,191 TIMES EXACTLY RIGHT AND THEN 70% 3234 02:17:27,191 --> 02:17:29,393 OF THE TIME GETTING THE TIME 3235 02:17:29,393 --> 02:17:31,562 CORRECT WITHIN ONE DAY AND SO 3236 02:17:31,562 --> 02:17:32,563 FORTH. 3237 02:17:32,563 --> 02:17:36,600 THE AREA UNDER THE CURVE IS 3238 02:17:36,600 --> 02:17:36,867 CAPTURED. 3239 02:17:36,867 --> 02:17:39,203 FINALLY WE MAPPED BACK TO 3240 02:17:39,203 --> 02:17:41,238 CONCORDANCE BECAUSE THAT'S WHAT 3241 02:17:41,238 --> 02:17:44,341 MAJORITY OF THIS COMMUNITY 3242 02:17:44,341 --> 02:17:46,577 FOCUSED ON BECAUSE TEMPORAL 3243 02:17:46,577 --> 02:17:48,612 RELATIONS, A MEASURE OF THAT 3244 02:17:48,612 --> 02:17:49,480 PERFORMANCE, CONCORDANCE, 3245 02:17:49,480 --> 02:17:53,384 WHETHER YOU'RE ORDERING EVENTS 3246 02:17:53,384 --> 02:17:54,652 THE SAME WAY, ANNOTATION, LARGE 3247 02:17:54,652 --> 02:17:55,552 LANGUAGE MODEL ANNOTATION. 3248 02:17:55,552 --> 02:17:58,022 SO THIS CAN BE SUMMARIZED IN THE 3249 02:17:58,022 --> 02:17:59,723 SUMMARY TABLE ON THE LEFT. 3250 02:17:59,723 --> 02:18:02,693 WE TRIED DIFFERENT MODELS. 3251 02:18:02,693 --> 02:18:06,497 THIS IS FOCUSING ON SEPSIS 3252 02:18:06,497 --> 02:18:10,000 COHORT, TIME SERIES, AND WE CAN 3253 02:18:10,000 --> 02:18:11,402 SEE THE LARGE FRONTIER MODELS 3254 02:18:11,402 --> 02:18:14,872 PERFORMED THE BEST, THERE ISN'T 3255 02:18:14,872 --> 02:18:20,444 A SINGLE BEST PERFORMER BUT 3256 02:18:20,444 --> 02:18:22,212 CONSUMER SMALLER ONES, PARAMETER 3257 02:18:22,212 --> 02:18:26,350 MODELS, THEY REALLY DON'T 3258 02:18:26,350 --> 02:18:26,684 PERFORM WELL. 3259 02:18:26,684 --> 02:18:28,585 PRETTY SHARP DEGRADATION TO THE 3260 02:18:28,585 --> 02:18:30,054 SMALLER MORE ACCESSIBLE MODELS. 3261 02:18:30,054 --> 02:18:32,289 SO BEING ABLE TO HAVE ACCESS TO 3262 02:18:32,289 --> 02:18:33,691 THESE FRONTIER MODELS IS REALLY 3263 02:18:33,691 --> 02:18:34,291 IMPORTANT. 3264 02:18:34,291 --> 02:18:37,962 ON THE RIGHT, WE ALSO PROVIDED 3265 02:18:37,962 --> 02:18:39,596 MANUAL ANNOTATIONS TO REVIEW OF 3266 02:18:39,596 --> 02:18:42,232 THE ANNOTATIONS WE GOT. 3267 02:18:42,232 --> 02:18:45,202 AND THIS ALLOWED US TO IDENTIFY, 3268 02:18:45,202 --> 02:18:47,237 YEAH, WE HAVE QUANTITATIVE 3269 02:18:47,237 --> 02:18:48,906 MEASURES FROM QUALITATIVE 3270 02:18:48,906 --> 02:18:50,574 PERSPECTIVE ARE ANNOTATIONS ALSO 3271 02:18:50,574 --> 02:18:52,509 OF HIGH QUALITY WHEN WE'RE 3272 02:18:52,509 --> 02:18:54,611 RELEASING CORPUS, DO WE THINK 3273 02:18:54,611 --> 02:18:57,548 THE ANNOTATIONS OF THE TIME ARE 3274 02:18:57,548 --> 02:18:57,781 STRONG. 3275 02:18:57,781 --> 02:19:01,051 SO WE CAN SEE, AGAIN, FRONTIER 3276 02:19:01,051 --> 02:19:02,086 MODELS HAVE ACCEPTABLE QUALITY, 3277 02:19:02,086 --> 02:19:05,889 HIGH PROPORTION OF THE TIME. 3278 02:19:05,889 --> 02:19:11,261 THE SMALLER MODELS REALLY DON'T 3279 02:19:11,261 --> 02:19:14,131 PROVIDE WHEN PERFORMING AT ALL. 3280 02:19:14,131 --> 02:19:16,967 SO JUST BACK TO HIGH LEVEL 3281 02:19:16,967 --> 02:19:21,605 AGAIN, WE'VE ILLUSTRATED 3282 02:19:21,605 --> 02:19:24,208 EXTRACTION FOR PubMed OPEN 3283 02:19:24,208 --> 02:19:26,910 ACCESS, ALSO IN PARALLEL FOR THE 3284 02:19:26,910 --> 02:19:29,646 DISCHARGE SUMMARIES, THE CORPUS 3285 02:19:29,646 --> 02:19:30,781 OF DISCHARGE SUMMARIES, 330,000 3286 02:19:30,781 --> 02:19:33,450 FROM A CRITICAL CARE DATASET. 3287 02:19:33,450 --> 02:19:37,087 AND IN THIS CASE WE HAVE ACCESS 3288 02:19:37,087 --> 02:19:40,824 TO BOTH THE TEXTUAL INFORMATION, 3289 02:19:40,824 --> 02:19:44,595 ALSO COMPLEMENTARY TABULAR DATA, 3290 02:19:44,595 --> 02:19:48,399 PROVIDING OPPORTUNITIES FOR 3291 02:19:48,399 --> 02:19:49,900 PATIENTS TO THINK ABOUT 3292 02:19:49,900 --> 02:19:55,906 MULTI-MODAL IN THE TABULARY 3293 02:19:55,906 --> 02:19:56,473 TEXTURE. 3294 02:19:56,473 --> 02:19:58,675 TO HIGHLIGHT A DOWNSTREAM USE, 3295 02:19:58,675 --> 02:20:02,079 WE COULD ASK CAN THESE MODELS 3296 02:20:02,079 --> 02:20:04,415 FORECAST WELL GIVEN THE TEXTUAL 3297 02:20:04,415 --> 02:20:04,915 TIME SERIES. 3298 02:20:04,915 --> 02:20:07,051 YOU MIGHT THINK MAYBE WE DON'T 3299 02:20:07,051 --> 02:20:09,319 NEED TO DO THIS, MAYBE LARGE 3300 02:20:09,319 --> 02:20:10,387 LANGUAGE MODELS OUT OF THE BOX 3301 02:20:10,387 --> 02:20:13,123 WILL BE ABLE TO ASSESS RISK FOR 3302 02:20:13,123 --> 02:20:13,323 THESE. 3303 02:20:13,323 --> 02:20:15,893 ONE WAY OF MEASURING IS ASKING, 3304 02:20:15,893 --> 02:20:18,729 OF THE EVENTS THAT DO OCCUR IN 3305 02:20:18,729 --> 02:20:20,931 THE 3306 02:20:20,931 --> 02:20:23,233 THE TIME SERIES WILL THE NEXT 3307 02:20:23,233 --> 02:20:28,172 EVENT OCCUR WITHIN ONE DAY, 3308 02:20:28,172 --> 02:20:29,273 PROMPTING THE ENGINEERING 3309 02:20:29,273 --> 02:20:31,575 EXERCISE, YOU GET A SCORE OF 3310 02:20:31,575 --> 02:20:31,875 .31. 3311 02:20:31,875 --> 02:20:34,945 YOU CAN DO A FAIR BIT BETTER, IF 3312 02:20:34,945 --> 02:20:36,847 YOU'RE GOING TO, YOU KNOW, BUILD 3313 02:20:36,847 --> 02:20:39,783 THIS INTERFACE, YOU'RE GOING TO 3314 02:20:39,783 --> 02:20:41,552 TRAIN MULTI-LAYERS ON THE OUTPUT 3315 02:20:41,552 --> 02:20:42,986 EMBEDDING OF THE LARGE LANGUAGE 3316 02:20:42,986 --> 02:20:45,055 MODEL, YOU CAN GET HIGHER POINT. 3317 02:20:45,055 --> 02:20:48,659 BETTER TO DO THIS ADDITIONAL 3318 02:20:48,659 --> 02:20:49,293 EXERCISE, NOT USE JUST LARGE 3319 02:20:49,293 --> 02:20:50,394 LANGUAGE MODEL STRAIGHT OUT OF 3320 02:20:50,394 --> 02:20:50,928 THE BOX. 3321 02:20:50,928 --> 02:20:51,595 THERE'S ALSO THIS QUESTION ABOUT 3322 02:20:51,595 --> 02:20:54,264 WHETHER OR NOT WE SHOULD BE 3323 02:20:54,264 --> 02:20:59,837 USING DECODER TYPE MODELS VERSUS 3324 02:20:59,837 --> 02:21:00,504 ENCODER. 3325 02:21:00,504 --> 02:21:03,640 MORE LIKE IN THE ChatGPT. 3326 02:21:03,640 --> 02:21:07,444 AND TURNS OUT THAT THE MORE 3327 02:21:07,444 --> 02:21:10,214 CONTEMPORARY VERSION OF THE BERT 3328 02:21:10,214 --> 02:21:11,982 MODELS HAVE HIGHER PERFORMANCE 3329 02:21:11,982 --> 02:21:14,952 CHARACTERISTICS, AND IF YOU USE 3330 02:21:14,952 --> 02:21:15,886 THE STANDARD MASKING PROCEDURE, 3331 02:21:15,886 --> 02:21:18,655 LET'S SAY USED FOR PREDICTION, 3332 02:21:18,655 --> 02:21:20,591 YOU GET THIS CHARACTERISTICS .33 3333 02:21:20,591 --> 02:21:23,060 BUT THE BEST PERFORMING SET OF 3334 02:21:23,060 --> 02:21:26,463 THE FORM OF THE MODEL IS USING 3335 02:21:26,463 --> 02:21:29,466 THE BERT-BASED MODELS OR 3336 02:21:29,466 --> 02:21:30,300 CONTEMPORARY VERSIONS, APPLYING 3337 02:21:30,300 --> 02:21:36,507 THIS INTERFACE TO GET PREDICTIVE 3338 02:21:36,507 --> 02:21:36,807 OUTPUT. 3339 02:21:36,807 --> 02:21:37,107 OKAY. 3340 02:21:37,107 --> 02:21:39,243 SO JUST KIND OF MOVING UP A 3341 02:21:39,243 --> 02:21:42,146 LEVEL AGAIN, WHAT ARE SOME 3342 02:21:42,146 --> 02:21:44,581 FUTURE DIRECTIONS FOR OUR GROUP? 3343 02:21:44,581 --> 02:21:46,783 IN THE TEMPORAL SUBGROUP AREA 3344 02:21:46,783 --> 02:21:48,185 WE'RE INTERESTED IN TWO FOCUS 3345 02:21:48,185 --> 02:21:49,486 AREAS. 3346 02:21:49,486 --> 02:21:51,288 WE'RE INTERESTED IN LOOKING AT 3347 02:21:51,288 --> 02:21:54,958 GLP-1 USERS WITH THE MAJOR 3348 02:21:54,958 --> 02:21:56,927 INCREASE IN POPULARITY OF THIS 3349 02:21:56,927 --> 02:21:59,696 MEDICATION, PARTICULARLY AWAY 3350 02:21:59,696 --> 02:22:01,865 FROM THE ORIGINAL INTENDED US, 3351 02:22:01,865 --> 02:22:03,300 SECONDARY TREATMENT FOR 3352 02:22:03,300 --> 02:22:03,800 DIABETES. 3353 02:22:03,800 --> 02:22:06,937 AND LOOKING AT SUBGROUPS TO 3354 02:22:06,937 --> 02:22:08,005 IDENTIFY ARE THERE PARTICULAR 3355 02:22:08,005 --> 02:22:10,007 INDIVIDUALS WHO RESPOND IN 3356 02:22:10,007 --> 02:22:11,141 PARTICULAR WAYS, PARTICULARLY 3357 02:22:11,141 --> 02:22:13,143 HELPFUL FROM A CARDIOVASCULAR 3358 02:22:13,143 --> 02:22:18,215 ASPECT OR PARTICULARLY, YOU 3359 02:22:18,215 --> 02:22:19,883 KNOW, DETRIMENTAL AND SO FORTH. 3360 02:22:19,883 --> 02:22:21,785 WE'VE GOT A NUMBER OF, YOU KNOW, 3361 02:22:21,785 --> 02:22:24,688 DATA RESOURCES THAT WE CAN APPLY 3362 02:22:24,688 --> 02:22:24,888 HERE. 3363 02:22:24,888 --> 02:22:27,124 WE WORK WITH "ALL OF US" DATA, 3364 02:22:27,124 --> 02:22:29,293 WORK WITH CMS DATA, AND WE ALSO 3365 02:22:29,293 --> 02:22:32,095 HAVE OUR TEXTUAL TIME SERIES 3366 02:22:32,095 --> 02:22:34,364 COHORTS, A COUPLE VIEWPOINTS TO 3367 02:22:34,364 --> 02:22:39,269 TRY TO TACKLE THIS QUESTION. 3368 02:22:39,269 --> 02:22:41,572 AND THEN IN COLLABORATION WITH 3369 02:22:41,572 --> 02:22:43,607 NIEHS WE WERE AWARDED THE NIH 3370 02:22:43,607 --> 02:22:46,009 DIRECTOR'S CHALLENGE AWARD LAST 3371 02:22:46,009 --> 02:22:47,811 YEAR, TO FOCUS ON LONGITUDINAL 3372 02:22:47,811 --> 02:22:51,014 AND MATERNAL HEALTH RISK AROUND 3373 02:22:51,014 --> 02:22:51,782 TIME OF CONCEPTION, 3374 02:22:51,782 --> 02:22:55,252 INVESTIGATING THAT WITH SOME OF 3375 02:22:55,252 --> 02:22:55,619 OUR TOOLS. 3376 02:22:55,619 --> 02:22:58,121 GOING BACK TO THE SERIES, WE'D 3377 02:22:58,121 --> 02:23:00,824 LIKE TO CONDUCT THIS PROCESS 3378 02:23:00,824 --> 02:23:02,125 BETTER, RECOGNIZING THAT 3379 02:23:02,125 --> 02:23:03,393 PATIENTS ARE COSTLY, SO WE'VE 3380 02:23:03,393 --> 02:23:05,462 GOT THAT SORT OF ACTIVE 3381 02:23:05,462 --> 02:23:06,663 LEARNING, HUMAN IN THE LOOP 3382 02:23:06,663 --> 02:23:08,699 PROCEDURE FOR TRYING TO GET THE 3383 02:23:08,699 --> 02:23:10,934 BEST ANNOTATIONS WE CAN POSSIBLY 3384 02:23:10,934 --> 02:23:13,737 GET WHILE, YOU KNOW, RESTRICTING 3385 02:23:13,737 --> 02:23:16,139 THE NUMBER OF ANNOTATIONS. 3386 02:23:16,139 --> 02:23:19,776 AND THEN GOING BACK TO THE 3387 02:23:19,776 --> 02:23:20,844 COMPLEMENTARITY AND TABULAR 3388 02:23:20,844 --> 02:23:26,116 DATA, THINKING ABOUT A 3389 02:23:26,116 --> 02:23:29,152 MULTI-MODAL ALIGNMENT OF DATA, 3390 02:23:29,152 --> 02:23:30,420 TO IMPROVE TEXTUAL TIME SERIES 3391 02:23:30,420 --> 02:23:32,256 ANNOTATIONS, ALSO TO GET BACK AT 3392 02:23:32,256 --> 02:23:35,259 THE QUESTION THAT WE HAVE 3393 02:23:35,259 --> 02:23:35,859 COMPLEMENTARY DATA, NOT USING 3394 02:23:35,859 --> 02:23:38,729 THEM AS MUCH AS WE COULD BE, 3395 02:23:38,729 --> 02:23:41,732 WHAT CAN WE SAY AND COMMUNICATE 3396 02:23:41,732 --> 02:23:43,100 ABOUT RISK, GIVEN WE HAVE ACCESS 3397 02:23:43,100 --> 02:23:45,068 TO BOTH IN A LONGITUDINAL 3398 02:23:45,068 --> 02:23:46,236 FASHION NOW. 3399 02:23:46,236 --> 02:23:49,439 AND THEN FINALLY GOING BACK TO 3400 02:23:49,439 --> 02:23:50,407 THE LEVEL OF EVIDENCE FIGURE 3401 02:23:50,407 --> 02:23:52,175 THAT I DESCRIBED AT THE VERY 3402 02:23:52,175 --> 02:23:56,313 BEGINNING OF THIS TALK, WE 3403 02:23:56,313 --> 02:23:58,315 FOCUSED ON OUR CASE REPORTS FOR 3404 02:23:58,315 --> 02:24:01,318 OUR TEXTUAL TIME SERIES 3405 02:24:01,318 --> 02:24:03,553 REPRESENTATION, HIGHER LEVELS OF 3406 02:24:03,553 --> 02:24:05,689 EVIDENCE ARE TYPICALLY 3407 02:24:05,689 --> 02:24:07,190 POPULATION LEVEL KIND OF PIECE 3408 02:24:07,190 --> 02:24:09,459 OF KNOWLEDGE, HOW CAN WE OVERLAY 3409 02:24:09,459 --> 02:24:10,460 THAT IN A TEMPORAL FASHION WITH 3410 02:24:10,460 --> 02:24:14,298 ALL THE CASE REPORTS WE HAVE 3411 02:24:14,298 --> 02:24:14,931 ANALYSIS FOR. 3412 02:24:14,931 --> 02:24:16,867 WE'D LIKE TO FIGURE OUT HOW TO 3413 02:24:16,867 --> 02:24:19,002 COMBINE THE TWO SO WE CAN 3414 02:24:19,002 --> 02:24:21,238 PROVIDE THE BEST, YOU KNOW, 3415 02:24:21,238 --> 02:24:22,739 EVIDENCE AND RECOMMENDATIONS 3416 02:24:22,739 --> 02:24:25,342 GOING FORWARD. 3417 02:24:25,342 --> 02:24:27,978 SO IN SUMMARY, TIME, WE BELIEVE 3418 02:24:27,978 --> 02:24:29,513 THE CORNERSTONE ELEMENT OF 3419 02:24:29,513 --> 02:24:30,247 HEALTH RISK, THEREFORE SOME 3420 02:24:30,247 --> 02:24:31,381 DECISION MAKING. 3421 02:24:31,381 --> 02:24:33,083 WE'VE EXPANDED OUR WORK INTO 3422 02:24:33,083 --> 02:24:35,185 TEXTUAL TIME SERIES, THIS 3423 02:24:35,185 --> 02:24:35,719 ENABLES MACHINE LEARNING 3424 02:24:35,719 --> 02:24:39,222 FORECASTING WITH CONTENT THAT 3425 02:24:39,222 --> 02:24:40,657 PROVIDERS ROUTINELY USE, BUT 3426 02:24:40,657 --> 02:24:43,260 OFTEN MODELS DON'T TAKE INTO 3427 02:24:43,260 --> 02:24:43,493 ACCOUNT. 3428 02:24:43,493 --> 02:24:46,997 AND THIS NEW WORLD OF LLMs WE 3429 02:24:46,997 --> 02:24:49,032 BELIEVE CLINICAL LARGE LANGUAGE 3430 02:24:49,032 --> 02:24:50,133 MODELS SHOULD BE IN A 3431 02:24:50,133 --> 02:24:51,501 QUANTITATIVE SENSE, THEY ARE NOT 3432 02:24:51,501 --> 02:24:53,970 THERE YET, WORKING TO CLOSE THAT 3433 02:24:53,970 --> 02:24:54,304 GAP. 3434 02:24:54,304 --> 02:25:00,243 SO THANK YOU FOR YOUR ATTENTION 3435 02:25:00,243 --> 02:25:02,713 THE GROUP WILL PRESENT RECENTLY 3436 02:25:02,713 --> 02:25:04,915 ACCEPTED WORK ON MEDICATION 3437 02:25:04,915 --> 02:25:08,118 ADHERENCE, AND I'M HAPPY TO TAKE 3438 02:25:08,118 --> 02:25:09,353 YOUR QUESTIONS. 3439 02:25:09,353 --> 02:25:10,020 THANKS. 3440 02:25:10,020 --> 02:25:11,188 >> THANKS A LOT, JEREMY. 3441 02:25:11,188 --> 02:25:13,857 NOW WE'RE GOING TO HAVE THE OPEN 3442 02:25:13,857 --> 02:25:14,491 QUESTION-AND-ANSWER SESSION 3443 02:25:14,491 --> 02:25:16,159 WHETHER WE CAN TAKE QUESTIONS 3444 02:25:16,159 --> 02:25:18,195 FROM BOARD MEMBERS OR THE 3445 02:25:18,195 --> 02:25:18,428 PUBLIC. 3446 02:25:18,428 --> 02:25:21,398 IT WILL BE 15 MINUTES, UNTIL 2: 3447 02:25:21,398 --> 02:25:23,033 35 AND THEN MOVED TO THE CLOSED 3448 02:25:23,033 --> 02:25:24,768 POSTER SESSION. 3449 02:25:24,768 --> 02:25:29,706 I SEE GRACIELA HAS A HAND UP. 3450 02:25:29,706 --> 02:25:29,973 GRACIELA? 3451 02:25:29,973 --> 02:25:31,274 >> SURE. 3452 02:25:31,274 --> 02:25:32,843 EXCELLENT WORK. 3453 02:25:32,843 --> 02:25:34,444 VERY, VERY INTERESTING. 3454 02:25:34,444 --> 02:25:36,246 I HAVE I GUESS ONE QUESTION 3455 02:25:36,246 --> 02:25:39,249 WRAPPED IN TWO, OR TWO QUESTIONS 3456 02:25:39,249 --> 02:25:41,418 WRAPPED IN ONE. 3457 02:25:41,418 --> 02:25:44,755 HAVE YOU CONSIDERED OR TRIED 3458 02:25:44,755 --> 02:25:46,490 YOUR MODELS ON ACTUAL CLINICAL 3459 02:25:46,490 --> 02:25:48,558 RECORDS, SOMETHING LIKE MIMIC OR 3460 02:25:48,558 --> 02:25:51,228 ANY OTHER CLINICAL RECORDS OTHER 3461 02:25:51,228 --> 02:25:52,763 THAN CASE REPORTS? 3462 02:25:52,763 --> 02:25:53,130 >> YES. 3463 02:25:53,130 --> 02:25:56,032 >> AND SECOND PART, WAS 3464 02:25:56,032 --> 02:25:57,267 BENCHMARKING, GOLD STANDARDS 3465 02:25:57,267 --> 02:25:59,736 THAT ARE AVAILABLE. 3466 02:25:59,736 --> 02:26:00,437 >> SURE. 3467 02:26:00,437 --> 02:26:07,110 YES, SO WE HAVE EXTRACTED THE 3468 02:26:07,110 --> 02:26:08,512 TEXTUAL TIME SERIES FOR UNIT 3469 02:26:08,512 --> 02:26:11,114 DISCHARGE SUMMARIES. 3470 02:26:11,114 --> 02:26:15,252 WE'VE DONE ANNOTATIONS OF THE 3471 02:26:15,252 --> 02:26:17,821 CASES, 23 CASES FROM MIMIC. 3472 02:26:17,821 --> 02:26:20,724 OVERLAP IS PARTIALLY WITH THE 3473 02:26:20,724 --> 02:26:23,794 i2b2 CORPUS FOCUSING ON 3474 02:26:23,794 --> 02:26:24,661 TEMPORAL RELATIONS, MIMIC 3475 02:26:24,661 --> 02:26:27,063 TWO-THREE, AN OLDER VERSION. 3476 02:26:27,063 --> 02:26:29,800 THEN ON MIMIC 4, MORE RECENT, A 3477 02:26:29,800 --> 02:26:34,004 LITTLE BIT MORE SANITIZED CASE 3478 02:26:34,004 --> 02:26:35,205 FOR DISCHARGE SUMMARIES. 3479 02:26:35,205 --> 02:26:37,607 >> SO YOU DID ANNOTATE THOSE? 3480 02:26:37,607 --> 02:26:38,241 THE MIMIC? 3481 02:26:38,241 --> 02:26:40,210 >> WE DID, YES, WE DID ANNOTATE 3482 02:26:40,210 --> 02:26:44,915 23 OF THOSE CASES AND WE'VE ALSO 3483 02:26:44,915 --> 02:26:47,050 EVALUATED, AND THOSE RESULTS ARE 3484 02:26:47,050 --> 02:26:51,788 FROM THE PAPER FROM THIS YEAR. 3485 02:26:51,788 --> 02:26:52,722 >> ALL RIGHT. 3486 02:26:52,722 --> 02:26:53,957 I WAS EXPECTING YOU WOULD 3487 02:26:53,957 --> 02:26:57,494 COMMENT ON HOW WELL IT PERFORMED 3488 02:26:57,494 --> 02:26:57,928 BUT LIKE -- 3489 02:26:57,928 --> 02:26:58,562 >> YEAH. 3490 02:26:58,562 --> 02:27:00,831 >> YOU DON'T HAVE THE EXACT 3491 02:27:00,831 --> 02:27:02,566 NUMBERS JUST LIKE -- I'M 3492 02:27:02,566 --> 02:27:02,799 CURIOUS. 3493 02:27:02,799 --> 02:27:04,734 >> SO I DON'T HAVE THE EXACT 3494 02:27:04,734 --> 02:27:07,270 NUMBERS IN THIS DECK. 3495 02:27:07,270 --> 02:27:09,306 THE PERFORMANCE DOES AGREE TO 3496 02:27:09,306 --> 02:27:11,608 SOME DEGREE. 3497 02:27:11,608 --> 02:27:14,511 SO THERE ARE -- DAYS REPORTS ARE 3498 02:27:14,511 --> 02:27:15,912 FORMATTED QUITE DIFFERENTLY THAN 3499 02:27:15,912 --> 02:27:17,647 THE DISCHARGE SUMMARIES. 3500 02:27:17,647 --> 02:27:20,383 THERE'S A LOT OF LIKE COPY/PASTE 3501 02:27:20,383 --> 02:27:23,220 IN THE DISCHARGE SUMMARIES, LIKE 3502 02:27:23,220 --> 02:27:28,124 TABLES WILL AUTOPOPULATE. 3503 02:27:28,124 --> 02:27:29,726 AND THOSE EVENTS CAN BE HARDER 3504 02:27:29,726 --> 02:27:31,161 TO CAPTURE FOR LARGE LANGUAGE 3505 02:27:31,161 --> 02:27:34,498 MODELS, SO WHAT WE SEE IS THE 3506 02:27:34,498 --> 02:27:37,000 EVENT MATCH RATE GOES DOWN 3507 02:27:37,000 --> 02:27:39,536 CONSIDERABLY, I THINK DOWN TO 3508 02:27:39,536 --> 02:27:42,072 LIKE 40 TO 50%. 3509 02:27:42,072 --> 02:27:45,242 AND THEN THE TEMPORAL ORDERING 3510 02:27:45,242 --> 02:27:46,843 ALSO DEGRADES TO SOME DEGREE. 3511 02:27:46,843 --> 02:27:48,545 I DON'T REMEMBER THE NUMBERS OFF 3512 02:27:48,545 --> 02:27:50,614 THE TOP OF MY HEAD. 3513 02:27:50,614 --> 02:27:52,482 >> NO, IT'S SORT OF WHAT I WAS 3514 02:27:52,482 --> 02:27:56,052 EXPECTING TO HEAR, BUT LIKE I'M 3515 02:27:56,052 --> 02:27:59,823 SURE THAT WAS MY QUESTION, I'LL 3516 02:27:59,823 --> 02:28:02,626 SAVE THIS FOR TIME FOR OTHER 3517 02:28:02,626 --> 02:28:04,594 PEOPLE. 3518 02:28:04,594 --> 02:28:04,928 >> RICHARD? 3519 02:28:04,928 --> 02:28:06,696 >> YEAH, I GUESS I'M ALLOWED TO 3520 02:28:06,696 --> 02:28:08,765 ASK QUESTIONS IN THE OPEN 3521 02:28:08,765 --> 02:28:08,999 SESSION. 3522 02:28:08,999 --> 02:28:10,433 SO, EARLIER WE HAD A 3523 02:28:10,433 --> 02:28:11,401 PRESENTATION FROM CLEM, AND AS 3524 02:28:11,401 --> 02:28:14,771 YOU KNOW HE WORKS ON POST-MARKET 3525 02:28:14,771 --> 02:28:16,606 SURVEILLANCE AND ONE OF HIS 3526 02:28:16,606 --> 02:28:22,145 FUTURE DIRECTIONS IS LOOKING AT 3527 02:28:22,145 --> 02:28:22,679 GLP-1 INHIBITORS. 3528 02:28:22,679 --> 02:28:25,115 I DON'T THINK HE HAS PLANS TO 3529 02:28:25,115 --> 02:28:26,116 INCORPORATE TEMPORAL RELATIONS 3530 02:28:26,116 --> 02:28:27,384 IN HIS EVALUATION. 3531 02:28:27,384 --> 02:28:29,419 HAVE YOU THOUGHT ABOUT MAYBE 3532 02:28:29,419 --> 02:28:32,255 PARTNERING UP WITH CLEM TO KIND 3533 02:28:32,255 --> 02:28:34,891 OF COMBINE FORCES ON THIS 3534 02:28:34,891 --> 02:28:35,158 QUESTION? 3535 02:28:35,158 --> 02:28:35,825 >> YEAH, I MEAN, THINK THAT 3536 02:28:35,825 --> 02:28:41,031 WOULD BE A REALLY EXCITING 3537 02:28:41,031 --> 02:28:41,331 OPPORTUNITY. 3538 02:28:41,331 --> 02:28:46,636 I WANT TO BE COGNIZANT OF LIKE 3539 02:28:46,636 --> 02:28:47,370 INSTITUTIONAL POLICIES, 3540 02:28:47,370 --> 02:28:48,038 INVESTIGATORS ARE OFTEN 3541 02:28:48,038 --> 02:28:52,242 ENCOURAGED, AS I UNDERSTAND IT, 3542 02:28:52,242 --> 02:28:54,511 TO, YOU KNOW, DEMONSTRATE THEIR 3543 02:28:54,511 --> 02:28:55,612 INDEPENDENCE AND SCIENTIFIC 3544 02:28:55,612 --> 02:28:56,146 AGENDA. 3545 02:28:56,146 --> 02:28:58,281 SO I THINK THERE'S A TON OF 3546 02:28:58,281 --> 02:28:58,715 OVERLAP. 3547 02:28:58,715 --> 02:29:00,183 I WOULD BE EXCITED TO WORK WITH 3548 02:29:00,183 --> 02:29:01,117 THEM. 3549 02:29:01,117 --> 02:29:04,487 I WANT TO MAKE SURE THAT'S 3550 02:29:04,487 --> 02:29:06,222 SOMETHING WE THINK IS STRATEGIC 3551 02:29:06,222 --> 02:29:09,526 FROM THE PERSPECTIVE, A GOOD 3552 02:29:09,526 --> 02:29:09,726 IDEA. 3553 02:29:09,726 --> 02:29:11,027 >> ONE THING, DO YOU REALLY 3554 02:29:11,027 --> 02:29:14,130 ESTABLISH YOURSELF AS KIND OF AN 3555 02:29:14,130 --> 02:29:15,398 EXPERT IN THIS TEMPORAL 3556 02:29:15,398 --> 02:29:17,834 COMPONENT TO THE ANALYSIS, AND 3557 02:29:17,834 --> 02:29:19,536 THAT'S NOT SOMETHING CLEM WAS 3558 02:29:19,536 --> 02:29:20,370 KNOWN FOR. 3559 02:29:20,370 --> 02:29:23,239 IF YOU WERE TO ADD THAT TO AND 3560 02:29:23,239 --> 02:29:25,075 COLLABORATE WITH HIM AND BRING 3561 02:29:25,075 --> 02:29:26,810 THAT COMPONENT IN I THINK YOU 3562 02:29:26,810 --> 02:29:27,944 WOULD BE RECOGNIZED FOR BRINGING 3563 02:29:27,944 --> 02:29:31,881 IT TO THE TABLE. 3564 02:29:31,881 --> 02:29:40,857 >> THANK YOU, GREAT IDEA. 3565 02:29:40,857 --> 02:29:45,528 >> OTHER QUESTIONS FOR JEREMY AT 3566 02:29:45,528 --> 02:29:47,230 THIS TIME? 3567 02:29:47,230 --> 02:29:50,567 >> YEAH, JUST WONDERING WITH THE 3568 02:29:50,567 --> 02:29:52,469 GLP-1, THINKING ABOUT THAT FROM 3569 02:29:52,469 --> 02:29:56,373 A TEMPORAL POINT OF VIEW, WHAT 3570 02:29:56,373 --> 02:29:58,274 TIME SCALE ARE YOU LOOKING AT 3571 02:29:58,274 --> 02:30:00,477 AND HOW LONG, YOU KNOW, GIVEN 3572 02:30:00,477 --> 02:30:03,947 THAT THESE ARE KIND OF COMING TO 3573 02:30:03,947 --> 02:30:05,982 THE MARKET, AND HAVE DIFFERENT 3574 02:30:05,982 --> 02:30:07,117 PRESCRIPTIONS FOR THEM, YOU 3575 02:30:07,117 --> 02:30:08,118 KNOW, WHAT TIME SCALE ARE YOU 3576 02:30:08,118 --> 02:30:10,687 LOOKING AT AND DO YOU PLAN ON 3577 02:30:10,687 --> 02:30:12,856 LOOKING AT KIND OF WHAT THE 3578 02:30:12,856 --> 02:30:22,065 REASON FOR GOING ON THESE IS? 3579 02:30:22,065 --> 02:30:23,967 >> YEAH, SO GLP-1s HAVE BEEN 3580 02:30:23,967 --> 02:30:26,569 AROUND OVER TWO DECADES. 3581 02:30:26,569 --> 02:30:29,406 BUT THE INDICATION FOR USE HAS 3582 02:30:29,406 --> 02:30:30,340 KIND OF SHIFTED DRAMATICALLY 3583 02:30:30,340 --> 02:30:32,742 OVER THE LAST COUPLE YEARS. 3584 02:30:32,742 --> 02:30:36,012 SO SINCE WE'RE INTERESTED IN 3585 02:30:36,012 --> 02:30:38,281 KIND OF THAT SHIFT, WE'RE 3586 02:30:38,281 --> 02:30:40,283 LOOKING AT RELATIVELY SHORTER 3587 02:30:40,283 --> 02:30:41,384 TIME, ON THE ORDER OF, YOU KNOW, 3588 02:30:41,384 --> 02:30:45,689 SIX MONTHS TO THREE YEARS, 3589 02:30:45,689 --> 02:30:46,823 ROUGHLY SPEAKING. 3590 02:30:46,823 --> 02:30:50,160 SUFFICIENT WHERE THERE WILL BE 3591 02:30:50,160 --> 02:30:53,663 SOME ADVERSE OUTCOMES, IN LARGE 3592 02:30:53,663 --> 02:30:53,963 POPULATIONS. 3593 02:30:53,963 --> 02:30:55,432 BUT NOT TO THE POINT WHERE WE 3594 02:30:55,432 --> 02:30:59,369 DON'T HAVE TO WAIT, YOU KNOW, 3595 02:30:59,369 --> 02:31:01,738 MANY YEARS TO ACTUALLY OBTAIN 3596 02:31:01,738 --> 02:31:03,440 THOSE OUTCOMES, ESPECIALLY WITH 3597 02:31:03,440 --> 02:31:08,044 THE RISE OF ITS USE FOR WEIGHT 3598 02:31:08,044 --> 02:31:08,278 LOSS. 3599 02:31:08,278 --> 02:31:15,251 WE ALREADY HAVE A SUBMITTED 3600 02:31:15,251 --> 02:31:20,790 MANUSCRIPT WHETHER WE -- GLP-1 3601 02:31:20,790 --> 02:31:21,891 FOCUSING ON AGAINST PROFILE, 3602 02:31:21,891 --> 02:31:22,826 PARTICULARLY SNP PROFILES. 3603 02:31:22,826 --> 02:31:25,028 IN THAT CASE WE'RE LOOKING, YOU 3604 02:31:25,028 --> 02:31:35,138 KNOW, AT ROUGHLY THAT TIME POIN. 3605 02:31:35,138 --> 02:31:37,107 >> I HAVE A COUPLE BIG PICTURE 3606 02:31:37,107 --> 02:31:39,142 QUESTIONS. 3607 02:31:39,142 --> 02:31:40,844 YOU MENTIONED A COUPLE TIMES 3608 02:31:40,844 --> 02:31:41,945 CAUSAL INFERENCE. 3609 02:31:41,945 --> 02:31:44,114 OBVIOUSLY TEMPORAL DIMENSION OF 3610 02:31:44,114 --> 02:31:45,048 YOUR WORK SUPPORTS CAUSAL 3611 02:31:45,048 --> 02:31:46,816 INFERENCE GIVEN THE TIME 3612 02:31:46,816 --> 02:31:48,585 PRECEDENCE OF THE RISK FACTORS 3613 02:31:48,585 --> 02:31:49,953 BEFORE THE OUTCOME. 3614 02:31:49,953 --> 02:31:51,321 BEYOND THE TEMPORAL DIMENSION, 3615 02:31:51,321 --> 02:31:52,522 ARE THERE OTHER METHODS YOU'RE 3616 02:31:52,522 --> 02:31:56,126 USING OR THINKING ABOUT USING TO 3617 02:31:56,126 --> 02:31:58,595 SUPPORT CAUSAL INFERENCE FROM 3618 02:31:58,595 --> 02:31:59,028 OBSERVATIONAL DATA? 3619 02:31:59,028 --> 02:32:01,598 >> YES, I THINK YOU'RE GOING TO 3620 02:32:01,598 --> 02:32:03,466 HEAR ABOUT THAT SPECIFICALLY 3621 02:32:03,466 --> 02:32:08,238 FROM SHARA AT THE POSTER 3622 02:32:08,238 --> 02:32:08,671 SESSION. 3623 02:32:08,671 --> 02:32:11,107 SO, WE THINK ABOUT PRIOR WORK 3624 02:32:11,107 --> 02:32:15,211 THAT FOCUSES ON MACHINE LEARNING 3625 02:32:15,211 --> 02:32:18,181 MODELS, OR CAUSAL INFERENCE, AND 3626 02:32:18,181 --> 02:32:20,116 CLASSICALLY YOU ADOPT BASICALLY 3627 02:32:20,116 --> 02:32:21,551 POTENTIAL OUTCOME FRAMEWORK FOR 3628 02:32:21,551 --> 02:32:22,786 OBSERVATIONAL DATA AND YOU'RE 3629 02:32:22,786 --> 02:32:24,387 GOING TO TRY TO DETERMINE 3630 02:32:24,387 --> 02:32:25,822 WHETHER OR NOT THE ASSUMPTIONS 3631 02:32:25,822 --> 02:32:29,893 THAT ARE REQUIRED FOR THAT 3632 02:32:29,893 --> 02:32:31,928 FRAMEWORK ARE, YOU KNOW, 3633 02:32:31,928 --> 02:32:32,428 JUSTIFIABLE. 3634 02:32:32,428 --> 02:32:34,097 AND ONE OF THE THINGS THAT YOU 3635 02:32:34,097 --> 02:32:37,300 CAN NEVER KNOW ABOUT IS 3636 02:32:37,300 --> 02:32:39,602 UNMEASURED CONFOUNDING, AND SO 3637 02:32:39,602 --> 02:32:42,405 ONE ARGUMENT ABOUT OUR WORK IS 3638 02:32:42,405 --> 02:32:43,473 THAT BECAUSE WE'RE EXTRACTING 3639 02:32:43,473 --> 02:32:44,374 THE BEST AVAILABLE INFORMATION 3640 02:32:44,374 --> 02:32:48,244 THAT WE CAN ABOUT THE PATIENTS, 3641 02:32:48,244 --> 02:32:52,282 IN A TEMPORAL WAY, THAT THIS IS 3642 02:32:52,282 --> 02:32:56,419 ACTUALLY IMPROVING OUR ABILITY 3643 02:32:56,419 --> 02:32:58,087 TO CHARACTERIZE THE COVARIATE, 3644 02:32:58,087 --> 02:33:01,758 WHICH GIVES YOU A BETTER CHANCE 3645 02:33:01,758 --> 02:33:03,827 AT MODELING THE UNMEASURED 3646 02:33:03,827 --> 02:33:06,563 CONFOUNDING THAT MAY NOT BE 3647 02:33:06,563 --> 02:33:08,264 CAPTURED IN SIMPLER 3648 02:33:08,264 --> 02:33:10,333 REPRESENTATIONS OF THE COVARIATE 3649 02:33:10,333 --> 02:33:11,000 PROFILE. 3650 02:33:11,000 --> 02:33:12,502 SO THAT'S ONE WAY. 3651 02:33:12,502 --> 02:33:15,705 AND THEN YOU'LL HEAR MORE FROM 3652 02:33:15,705 --> 02:33:17,574 SHARA BUT, YOU KNOW, THERE ARE 3653 02:33:17,574 --> 02:33:22,245 SOME OPEN QUESTIONS WITH RESPECT 3654 02:33:22,245 --> 02:33:24,881 TO HOW CAUSAL INFERENCE METHODS 3655 02:33:24,881 --> 02:33:27,250 CAN BE APPLIED IN SURVIVAL 3656 02:33:27,250 --> 02:33:28,151 ANALYSIS SETTINGS. 3657 02:33:28,151 --> 02:33:30,987 AND SO WE'VE GOT A LINE OF WORK 3658 02:33:30,987 --> 02:33:34,457 THAT TRIES TO DEMONSTRATE WHAT 3659 02:33:34,457 --> 02:33:35,625 ARE BEST PRACTICES WHEN WE 3660 02:33:35,625 --> 02:33:38,361 COMBINE THESE TWO AREAS OF STUDY 3661 02:33:38,361 --> 02:33:42,298 AND YOU'LL SEE THAT IN MORE 3662 02:33:42,298 --> 02:33:42,532 DETAIL. 3663 02:33:42,532 --> 02:33:46,135 >> I LOOK FORWARD TO THAT. 3664 02:33:46,135 --> 02:33:46,369 THANKS. 3665 02:33:46,369 --> 02:33:46,636 GRACIELA? 3666 02:33:46,636 --> 02:33:47,570 >> JUST A POINT OF 3667 02:33:47,570 --> 02:33:49,906 CLARIFICATION ABOUT YOUR WORK ON 3668 02:33:49,906 --> 02:33:51,207 SEPSIS. 3669 02:33:51,207 --> 02:33:53,142 THE OUTCOME THAT YOU WERE 3670 02:33:53,142 --> 02:33:55,345 PREDICTING OR MEASURING IS ONLY 3671 02:33:55,345 --> 02:33:58,581 SURVIVAL OR DO YOU DO ANY 3672 02:33:58,581 --> 02:33:58,815 OTHERS? 3673 02:33:58,815 --> 02:34:01,551 OR PRESENCE OR ABSENCE OF 3674 02:34:01,551 --> 02:34:02,118 SEPSIS? 3675 02:34:02,118 --> 02:34:03,753 OR IT WASN'T CLEAR TO MY WHEN 3676 02:34:03,753 --> 02:34:05,188 YOU PRESENTED, WHAT WAS IT THAT 3677 02:34:05,188 --> 02:34:07,223 YOU WERE TRYING TO PREDICT IN 3678 02:34:07,223 --> 02:34:07,624 THE EVENT? 3679 02:34:07,624 --> 02:34:08,258 >> SURE. 3680 02:34:08,258 --> 02:34:10,660 YES, SO MORTALITY IS THE PRIMARY 3681 02:34:10,660 --> 02:34:11,895 OUTCOME. 3682 02:34:11,895 --> 02:34:14,764 THERE ARE ALSO SECONDARY 3683 02:34:14,764 --> 02:34:18,935 OUTCOMES OF MULTI-ORGAN FAILURE 3684 02:34:18,935 --> 02:34:19,903 AND MECHANICAL VENTILATION. 3685 02:34:19,903 --> 02:34:22,405 DEPENDING WHICH WORK YOU'RE 3686 02:34:22,405 --> 02:34:26,509 REFERRAL TO IT WILL BE A 3687 02:34:26,509 --> 02:34:27,744 MIXTURE, WE'RE LOOKING 3688 02:34:27,744 --> 02:34:30,747 DOWNSTREAM, NOT TRYING TO DETECT 3689 02:34:30,747 --> 02:34:30,980 SEPSIS. 3690 02:34:30,980 --> 02:34:32,282 >> IS IT WITHIN A CERTAIN PERIOD 3691 02:34:32,282 --> 02:34:32,749 OF TIME? 3692 02:34:32,749 --> 02:34:37,053 BECAUSE IF YOU WERE USING DATA 3693 02:34:37,053 --> 02:34:40,256 IN THE FIRST SIX HOURS I THINK 3694 02:34:40,256 --> 02:34:45,161 YOU SAID, WAS THERE A WINDOW IN 3695 02:34:45,161 --> 02:34:46,095 YOUR LONGITUDINAL PREDICTION? 3696 02:34:46,095 --> 02:34:50,199 >> YES, IT VARIES. 3697 02:34:50,199 --> 02:34:52,735 IT COULD BE WITH LIKE INPATIENT 3698 02:34:52,735 --> 02:34:54,971 MORTALITY, IT COULD BE 30-DAY 3699 02:34:54,971 --> 02:34:58,207 MORTALITY, IT COULD BE 90-DAY 3700 02:34:58,207 --> 02:35:02,412 MORTALITY. 3701 02:35:02,412 --> 02:35:02,679 >> OKAY. 3702 02:35:02,679 --> 02:35:03,646 ALL RIGHT. 3703 02:35:03,646 --> 02:35:04,747 AND IF NOBODY ELSE HAS OTHER 3704 02:35:04,747 --> 02:35:07,650 QUESTIONS I HAVE ONE MORE, WHICH 3705 02:35:07,650 --> 02:35:09,252 IS ABOUT SHARED TASKS. 3706 02:35:09,252 --> 02:35:11,387 HAVE YOU ORGANIZED OR 3707 02:35:11,387 --> 02:35:13,823 PARTICIPATED, OR ORGANIZED IS MY 3708 02:35:13,823 --> 02:35:15,124 QUESTION, ACTUALLY, SHARED TASKS 3709 02:35:15,124 --> 02:35:19,762 WITH DATA THAT YOU HAVE 3710 02:35:19,762 --> 02:35:20,430 ANNOTATED AND CREATED? 3711 02:35:20,430 --> 02:35:22,265 >> NOT YET. 3712 02:35:22,265 --> 02:35:28,304 SO WE ARE IN PREPARATION FOR A 3713 02:35:28,304 --> 02:35:29,739 SUBMISSION FOR BENCHMARKS VENUE 3714 02:35:29,739 --> 02:35:31,207 COMING IN A MONTH, AND SO PART 3715 02:35:31,207 --> 02:35:34,844 OF THAT RELEASE WAS GOING TO BE, 3716 02:35:34,844 --> 02:35:36,612 YOU KNOW, GATHER PEOPLE TO 3717 02:35:36,612 --> 02:35:37,747 ITERATE AND FIGURE OUT HOW WELL 3718 02:35:37,747 --> 02:35:38,881 THEY CAN DO. 3719 02:35:38,881 --> 02:35:40,116 >> ALL RIGHT. 3720 02:35:40,116 --> 02:35:43,786 JUST FOR THE RECORD, BIOCREATIVE 3721 02:35:43,786 --> 02:35:45,688 IS ORGANIZED, I THINK THIS WILL 3722 02:35:45,688 --> 02:35:49,826 BE PERFECT FOR THE FUTURE 3723 02:35:49,826 --> 02:35:50,126 ITERATIONS. 3724 02:35:50,126 --> 02:35:51,527 WE'RE RUNNING THE NEXT ONE 3725 02:35:51,527 --> 02:35:53,529 ALREADY AND WE HAVE THE SHARED 3726 02:35:53,529 --> 02:35:56,966 TASK, BUT IT WILL BE GREAT TO 3727 02:35:56,966 --> 02:35:58,901 SEE YOU ON THAT. 3728 02:35:58,901 --> 02:35:59,936 OR OTHERS. 3729 02:35:59,936 --> 02:36:02,372 SHARED TASKS PLAY A REALLY 3730 02:36:02,372 --> 02:36:04,240 PIVOTAL ROLE IN MOTIVATING OR 3731 02:36:04,240 --> 02:36:06,776 PEOPLE AND PREPARING OUR WORK 3732 02:36:06,776 --> 02:36:07,744 AND SO ON. 3733 02:36:07,744 --> 02:36:13,850 >> YEAH, SOUNDS LIKE A GREAT 3734 02:36:13,850 --> 02:36:14,450 OPPORTUNITY. 3735 02:36:14,450 --> 02:36:15,918 >> JEREMY, YOU GAVE AN 3736 02:36:15,918 --> 02:36:17,954 INTERESTING PREVIEW OF YOUR 3737 02:36:17,954 --> 02:36:21,391 LARGE LANGUAGE MODEL TOWARDS THE 3738 02:36:21,391 --> 02:36:27,230 END, LLMs ARE GOODED A MAKING 3739 02:36:27,230 --> 02:36:29,966 PREDICTIONS, USING LATENT SPACE 3740 02:36:29,966 --> 02:36:30,600 TO MAKE PREDICTIONS, WONDERING 3741 02:36:30,600 --> 02:36:32,902 IF YOU HAVE INTEREST OR PLANS IN 3742 02:36:32,902 --> 02:36:35,071 TRYING TO UNPACK OR DECODE THAT 3743 02:36:35,071 --> 02:36:37,173 LATENT SPACE AND FIGURE OUT THE 3744 02:36:37,173 --> 02:36:42,245 RISK FACTORS OF INTEREST, WHAT 3745 02:36:42,245 --> 02:36:44,580 IS THE INFORMATION CLINICS ARE 3746 02:36:44,580 --> 02:36:45,648 COLLECTING MOST RELEVANT TO YOUR 3747 02:36:45,648 --> 02:36:46,916 OUTCOMES OF INTEREST. 3748 02:36:46,916 --> 02:36:50,053 >> THAT'S A VERY INTERESTING 3749 02:36:50,053 --> 02:36:53,523 QUESTION, A REALLY IMPORTANT 3750 02:36:53,523 --> 02:36:53,790 QUESTION. 3751 02:36:53,790 --> 02:36:55,558 A Ph.D. IN INTERPRETIVE 3752 02:36:55,558 --> 02:36:57,193 MACHINE LEARNING FOR HEALTH 3753 02:36:57,193 --> 02:36:59,495 CARE, SO THAT'S SOMETHING THAT 3754 02:36:59,495 --> 02:37:02,965 HE COULD DEFINITELY LOOK TO 3755 02:37:02,965 --> 02:37:03,433 TACKLE. 3756 02:37:03,433 --> 02:37:05,968 WE HAVE WORKED TRYING TO DETECT 3757 02:37:05,968 --> 02:37:07,236 LIKE THE HIGH-LEVEL CONCEPTS 3758 02:37:07,236 --> 02:37:09,939 THAT CAN BE USED IN AND 3759 02:37:09,939 --> 02:37:12,842 TRANSPORT BETTER WHEN MAKING 3760 02:37:12,842 --> 02:37:14,777 PREDICTIONS FROM, YOU KNOW, 3761 02:37:14,777 --> 02:37:17,980 ESSENTIALLY BLACK BOX TYPE 3762 02:37:17,980 --> 02:37:21,851 MODELS, HAVEN'T APPLIED THEM TO 3763 02:37:21,851 --> 02:37:23,953 EMBEDDING SPACINGS OF LARGE 3764 02:37:23,953 --> 02:37:25,655 LANGUAGE MODELS, THERE'S WORK IN 3765 02:37:25,655 --> 02:37:27,090 FROM INTERROGATING EMBEDDING 3766 02:37:27,090 --> 02:37:28,391 SPACE OF THESE MODELS. 3767 02:37:28,391 --> 02:37:32,562 SO, YEAH, I THINK IT'S AN 3768 02:37:32,562 --> 02:37:34,964 IMPORTANT QUESTION. 3769 02:37:34,964 --> 02:37:36,165 WE HAVEN'T JUMPED INTO THAT 3770 02:37:36,165 --> 02:37:38,167 SPACE WITH OUR CURRENT PROJECT, 3771 02:37:38,167 --> 02:37:40,536 BUT IT'S ON THE -- WE'RE 3772 02:37:40,536 --> 02:37:41,237 DEFINITELY THINKING BIT, 3773 02:37:41,237 --> 02:37:42,839 THINKING WE HAVE AN ANGLE THAT 3774 02:37:42,839 --> 02:37:44,841 WE THINK, YOU KNOW, WILL HELP US 3775 02:37:44,841 --> 02:37:48,344 OR LIKE WHERE WE CAN DO IT 3776 02:37:48,344 --> 02:37:49,545 BETTER AND SHOW THE COMMUNITY 3777 02:37:49,545 --> 02:37:51,114 THAT IT'S ALREADY THINKING ABOUT 3778 02:37:51,114 --> 02:37:54,984 THAT, YOU KNOW, WHAT WE CAN 3779 02:37:54,984 --> 02:37:55,451 OFFER. 3780 02:37:55,451 --> 02:37:56,185 >> GOOD. 3781 02:37:56,185 --> 02:37:56,419 THANKS. 3782 02:37:56,419 --> 02:37:57,954 I DON'T SEE ANY OTHER HANDS UP. 3783 02:37:57,954 --> 02:38:00,957 DO WE HAVE ANY MORE QUESTIONS 3784 02:38:00,957 --> 02:38:04,193 FOR JEREMY BEFORE WE MOVE TO THE 3785 02:38:04,193 --> 02:38:12,034 CLOSED POSTER SESSION? 3786 02:38:12,034 --> 02:38:13,169 OKAY. 3787 02:38:13,169 --> 02:38:23,169 JEREMY, THANKS A LOT.