1
00:00:05,000 --> 00:00:07,520
I WANT TO WELCOME YOU TO THE

2
00:00:07,520 --> 00:00:09,160
WEDNESDAY AFTERNOON LECTURE

3
00:00:09,160 --> 00:00:09,480
SERIES.

4
00:00:09,480 --> 00:00:11,040
I'M A GRADUATE STUDENT AT

5
00:00:11,040 --> 00:00:12,640
GEORGETOWN UNIVERSITY AND I'M

6
00:00:12,640 --> 00:00:14,640
DOING MY RESEARCH HERE AT THE

7
00:00:14,640 --> 00:00:16,000
NATIONAL HUMAN GENOME RESEARCH

8
00:00:16,000 --> 00:00:17,080
INSTITUTE.

9
00:00:17,080 --> 00:00:18,800
AND YEAH, I'M REALLY EXCITED

10
00:00:18,800 --> 00:00:20,200
TODAY TO BE INTRODUCING

11
00:00:20,200 --> 00:00:21,880
DR. DAVID BAKER, WHO'S HERE WITH

12
00:00:21,880 --> 00:00:23,720
US TODAY.

13
00:00:23,720 --> 00:00:25,240
DAVID GOT HIS START WITH A

14
00:00:25,240 --> 00:00:27,200
PH.D. IN BIOCHEMISTRY WORKING

15
00:00:27,200 --> 00:00:32,720
WITH RANDY SHEKMAN AT BERKELEY

16
00:00:32,720 --> 00:00:38,440
AND HE IS NOW PROFESSOR AT THE

17
00:00:38,440 --> 00:00:40,760
UNIVERSITY OF WASHINGTON, HOWARD

18
00:00:40,760 --> 00:00:42,800
HUGHES MEDICAL INVESTIGATOR, AND

19
00:00:42,800 --> 00:00:44,480
INSTITUTE FOR PROTEIN DESIGN.

20
00:00:44,480 --> 00:00:46,240
IN 2021, WELL, MAYBE I SHOULD

21
00:00:46,240 --> 00:00:48,080
JUST SAY LIKE 2021 WAS A HUGE

22
00:00:48,080 --> 00:00:50,480
YEAR FOR THE FIELD AT LARGE BUT

23
00:00:50,480 --> 00:00:51,600
ESPECIALLY FOR THE BAKER GROUP

24
00:00:51,600 --> 00:00:54,280
WITH THE DEVELOPMENT AND RELEASE

25
00:00:54,280 --> 00:01:00,880
OF ROSETAFOLD AND TOGETHER

26
00:01:00,880 --> 00:01:01,640
COMPLETELY TRANSFORMED THE FIELD

27
00:01:01,640 --> 00:01:03,320
WHICH HAPPENS I THINK PRETTY

28
00:01:03,320 --> 00:01:04,280
RARELY, BUT YEAH.

29
00:01:04,280 --> 00:01:07,240
IT WAS PRETTY EXCITING.

30
00:01:07,240 --> 00:01:08,560
IT GIVES ME GOOSE BUMPS.

31
00:01:08,560 --> 00:01:10,960
IT WAS RECOGNIZED, THE TWO WERE

32
00:01:10,960 --> 00:01:12,800
RECOGNIZED FOR THE BREAKTHROUGH

33
00:01:12,800 --> 00:01:14,520
OF THE YEAR IN 2021 BY THE

34
00:01:14,520 --> 00:01:17,400
JOURNAL SCIENCE.

35
00:01:17,400 --> 00:01:18,440
DAVID'S SCIENTIFIC METRICS ARE

36
00:01:18,440 --> 00:01:19,360
COMPLETELY OFF THE CHARTS,

37
00:01:19,360 --> 00:01:22,520
WHETHER YOU LOOK AT

38
00:01:22,520 --> 00:01:23,680
PUBLICATIONS, COMPANIES FOUNDED,

39
00:01:23,680 --> 00:01:25,600
ALL THESE NUMBERS, THEY'RE JUST

40
00:01:25,600 --> 00:01:27,240
HUGE NUMBERS AND I DON'T WANT TO

41
00:01:27,240 --> 00:01:29,200
DOWNPLAY THAT BUT WHAT I WANT TO

42
00:01:29,200 --> 00:01:29,960
EMPHASIZE MAYBE IS THE

43
00:01:29,960 --> 00:01:30,960
DEDICATION HE HAS TO THE

44
00:01:30,960 --> 00:01:34,120
WELL-BEING OF HIS TRAINEES,

45
00:01:34,120 --> 00:01:35,120
WHICH IS A SMALL ARMY.

46
00:01:35,120 --> 00:01:36,560
THE REASON I WANT TO EMPHASIZE

47
00:01:36,560 --> 00:01:37,760
THIS IS THESE ARE THE FOLKS THAT

48
00:01:37,760 --> 00:01:39,040
ARE GOING TO CONTINUE TO

49
00:01:39,040 --> 00:01:39,960
REVOLUTIONIZE THE FIELD FOR

50
00:01:39,960 --> 00:01:43,120
DECADES TO COME IN WAYS WE CAN'T

51
00:01:43,120 --> 00:01:43,760
EVEN IMAGINE.

52
00:01:43,760 --> 00:01:44,840
AND THE LAST THING I WANT TO

53
00:01:44,840 --> 00:01:47,800
TOUCH ON HERE IS, IT'S REALLY A

54
00:01:47,800 --> 00:01:49,440
SPECIAL HONOR FOR ME TO BE

55
00:01:49,440 --> 00:01:50,800
INTRODUCING DAVID TODAY.

56
00:01:50,800 --> 00:01:53,400
HE'S HAD A PRETTY -- EXCUSE ME.

57
00:01:53,400 --> 00:01:56,560
HE'S HAD A SIGNIFICANT IMPACT ON

58
00:01:56,560 --> 00:01:58,200
MY CAREER AT LEAST AS FAR AS

59
00:01:58,200 --> 00:02:01,040
INSPIRING ME TO GET INVOLVED IN

60
00:02:01,040 --> 00:02:01,680
BIOMEDICAL RESEARCH.

61
00:02:01,680 --> 00:02:02,680
WHO DOESN'T WANT TO MAKE

62
00:02:02,680 --> 00:02:03,240
PROTEINS?

63
00:02:03,240 --> 00:02:04,120
IT'S PRETTY ENTICING.

64
00:02:04,120 --> 00:02:05,320
ALSO I WANT TO MENTION TOO,

65
00:02:05,320 --> 00:02:06,840
DAVID IS NOT AFRAID TO THINK

66
00:02:06,840 --> 00:02:07,480
BIG.

67
00:02:07,480 --> 00:02:08,720
HE'S BEEN DEDICATING HIS CAREER

68
00:02:08,720 --> 00:02:10,040
TO A PRETTY DARN DIFFICULT

69
00:02:10,040 --> 00:02:11,320
PROBLEM THAT A LOT OF FOLKS

70
00:02:11,320 --> 00:02:12,240
DIDN'T THINK COULD BE ACHIEVED

71
00:02:12,240 --> 00:02:15,720
IN THIS LIFETIME, AND SO I

72
00:02:15,720 --> 00:02:17,560
REALLY ADMIRE BOTH HIS

73
00:02:17,560 --> 00:02:19,640
PERSISTENCE AND HIS ABILITY TO

74
00:02:19,640 --> 00:02:21,600
APPLY NEW NOVEL METHODS TO TOUGH

75
00:02:21,600 --> 00:02:22,160
PROBLEMS.

76
00:02:22,160 --> 00:02:23,560
SO WITH THAT SAID, THANK YOU,

77
00:02:23,560 --> 00:02:25,040
DAVID, FOR BEING HERE IN PERSON

78
00:02:25,040 --> 00:02:25,440
TODAY.

79
00:02:25,440 --> 00:02:27,280
I'M REALLY EXCITED FOR YOUR TALK

80
00:02:27,280 --> 00:02:30,560
TITLED "THE COMING OF AGE 

81
00:02:30,560 --> 00:02:36,240
OF DE NOVO 

82
00:02:36,240 --> 00:02:38,680
>>TODAY I WANT TO TELL ABOUT DE

83
00:02:38,680 --> 00:02:40,520
NOVO PROTEIN DESIGN, SO AS YOU

84
00:02:40,520 --> 00:02:43,480
KNOW, PROTEINS AND NATURE CARRY

85
00:02:43,480 --> 00:02:45,240
OUT A REALLY WIDE RANGE OF

86
00:02:45,240 --> 00:02:47,520
IMPORTANT FUNCTIONS AND DE NOVO

87
00:02:47,520 --> 00:02:48,400
DEVINE IS BASICALLY THE PROBLEM

88
00:02:48,400 --> 00:02:49,920
OF MAKING BRAND NEW PROTEINS

89
00:02:49,920 --> 00:02:51,040
WITH NEW FUNCTIONS TO SOLVE

90
00:02:51,040 --> 00:02:52,440
PROBLEMS THAT WEREN'T REALLY

91
00:02:52,440 --> 00:02:55,200
AROUND DURING NATURAL EVOLUTION,

92
00:02:55,200 --> 00:02:56,440
WHEN ALL THE NATURAL PROTEINS

93
00:02:56,440 --> 00:02:57,480
CAME ABOUT.

94
00:02:57,480 --> 00:03:00,720
AND SO FOR MANY YEARS, WE

95
00:03:00,720 --> 00:03:01,520
APPROACH THIS PROBLEM SORT OF

96
00:03:01,520 --> 00:03:03,000
USING A PHYSICALLY-BASED MODEL

97
00:03:03,000 --> 00:03:04,560
WHERE BASED ON THE IDEA THAT

98
00:03:04,560 --> 00:03:06,000
PROTEINS FOLD TO THEIR LOWEST

99
00:03:06,000 --> 00:03:06,880
ENERGY STATES.

100
00:03:06,880 --> 00:03:08,840
SO IF WE WANT TO DESIGN A

101
00:03:08,840 --> 00:03:10,400
PROTEIN FOLD TO A GIVEN

102
00:03:10,400 --> 00:03:11,800
STRUCTURE AND HAVE A GIVEN

103
00:03:11,800 --> 00:03:14,120
FUNCTION, WE SEARCHED FOR AN

104
00:03:14,120 --> 00:03:15,520
AMINO ACID SEQUENCE WHOSE LOWEST

105
00:03:15,520 --> 00:03:17,280
ENERGY STATE WAS THAT STRUCTURE.

106
00:03:17,280 --> 00:03:19,000
NOW MORE RECENTLY, AS YOU'RE ALL

107
00:03:19,000 --> 00:03:20,880
AWARE, DEEP LEARNING HAS BECOME

108
00:03:20,880 --> 00:03:21,840
INCREASINGLY POWERFUL, AND SO

109
00:03:21,840 --> 00:03:24,720
NOW WE CAN USE DEEP LEARNING

110
00:03:24,720 --> 00:03:26,440
METHODS TO DESIGN PROTEINS NOT

111
00:03:26,440 --> 00:03:28,520
USING SORT OF A PHYSICALLY BASED

112
00:03:28,520 --> 00:03:29,960
MODEL BUT USING A DEEP NEURAL

113
00:03:29,960 --> 00:03:30,400
NETWORK.

114
00:03:30,400 --> 00:03:31,840
AND I'LL TALK ABOUT THAT IN THE

115
00:03:31,840 --> 00:03:35,080
SECOND PART OF MY TALK.

116
00:03:35,080 --> 00:03:36,440
I WANT TO SORT OF -- THERE'S

117
00:03:36,440 --> 00:03:38,040
ALSO A THEME OF PANDEMIC

118
00:03:38,040 --> 00:03:39,360
PREPAREDNESS THAT RUNS THROUGH

119
00:03:39,360 --> 00:03:45,920
MY TALK, SO WE'VE BEEN DESIGNING

120
00:03:45,920 --> 00:03:49,840
BOTH VACCINES AND ANTIVIRALS,

121
00:03:49,840 --> 00:03:51,080
AND IN THE FIRST PART OF MY

122
00:03:51,080 --> 00:03:52,920
TALK, I'LL SHOW YOU HOW WE'VE

123
00:03:52,920 --> 00:03:56,760
BEEN ABLE TO MAKE QUITE A NUMBER

124
00:03:56,760 --> 00:03:59,720
OF ANTIVIRALS THAT COULD BE

125
00:03:59,720 --> 00:04:01,880
USEFUL AGAINST FUTURE PANDEMICS,

126
00:04:01,880 --> 00:04:03,520
AND THEN -- AND THAT'S WHAT WE

127
00:04:03,520 --> 00:04:05,320
CALL JUST IN CASE, BECAUSE

128
00:04:05,320 --> 00:04:06,840
THAT'S SORT OF ANTICIPATING

129
00:04:06,840 --> 00:04:07,800
WHERE THE NEXT PROBLEM IS GOING

130
00:04:07,800 --> 00:04:09,000
TO COME FROM, BUT WE ALSO WOULD

131
00:04:09,000 --> 00:04:10,760
LIKE TO HAVE REALLY FAST METHODS

132
00:04:10,760 --> 00:04:13,800
FOR MAKING THERAPEUTICS IN THE

133
00:04:13,800 --> 00:04:15,480
EVENT OF A NEW PANDEMIC

134
00:04:15,480 --> 00:04:16,680
OUTBREAK, AND WHEN I GET TO THE

135
00:04:16,680 --> 00:04:17,880
DEEP LEARNING PART, I'LL SHOW

136
00:04:17,880 --> 00:04:19,960
YOU HOW NOW WE CAN DO EVERYTHING

137
00:04:19,960 --> 00:04:21,240
MUCH, MUCH FASTER THAN WE COULD

138
00:04:21,240 --> 00:04:23,000
BEFORE.

139
00:04:23,000 --> 00:04:25,840
SO IT'S AN EXCITING TIME FOR US.

140
00:04:25,840 --> 00:04:28,600
WE HAVE OUR FIRST

141
00:04:28,600 --> 00:04:29,080
COMPUTATIONALLY DESIGNED

142
00:04:29,080 --> 00:04:30,640
MEDICINE.

143
00:04:30,640 --> 00:04:32,080
THIS IS SOME NUMBER OF YEARS

144
00:04:32,080 --> 00:04:35,040
AGO, WE -- OOPS, LET'S SEE.

145
00:04:35,040 --> 00:04:38,880
WE DESIGNED SELF ASSEMBLING

146
00:04:38,880 --> 00:04:39,440
NANOPARTICLES.

147
00:04:39,440 --> 00:04:40,520
MY COLLEAGUE KNEEL KING, THOSE

148
00:04:40,520 --> 00:04:46,520
NEIL KING PUT THERECEPTOR-BINDIE

149
00:04:46,520 --> 00:04:47,520
CORONAVIRUS ON IT AND IT TURNS

150
00:04:47,520 --> 00:04:49,480
OUT TO ELICIT A REALLY, REALLY

151
00:04:49,480 --> 00:04:50,560
STRONG IMMUNE RESPONSE, QUITE A

152
00:04:50,560 --> 00:04:52,320
BIT STRONGER THAN THE MRNA

153
00:04:52,320 --> 00:04:52,880
VACCINES.

154
00:04:52,880 --> 00:04:54,360
IT'S ALSO VERY, VERY STABLE SO

155
00:04:54,360 --> 00:04:56,520
IT DOESN'T REQUIRE A COLD CHAIN.

156
00:04:56,520 --> 00:04:59,200
SO THIS IS NOW IN USE IN KOREA

157
00:04:59,200 --> 00:05:03,360
AND HOPEFULLY SOON ELSEWHERE.

158
00:05:03,360 --> 00:05:04,720
SO THE FIRST PART OF MY TALK

159
00:05:04,720 --> 00:05:06,240
WILL BE ABOUT DESIGN OF BINDING

160
00:05:06,240 --> 00:05:09,160
PROTEINS.

161
00:05:09,160 --> 00:05:10,360
AND SO WE'VE DEVELOPED GENERAL

162
00:05:10,360 --> 00:05:12,240
METHODS AND THIS IS USING THE

163
00:05:12,240 --> 00:05:15,840
PHYSICALLY-BASED APPROACH TO

164
00:05:15,840 --> 00:05:17,280
DESIGN PROTEINS THAT ARE SHAPED

165
00:05:17,280 --> 00:05:24,720
AND CHEMICALLY CL CL COMPLIMENTY

166
00:05:24,720 --> 00:05:26,760
TARGETS LIKE THE

167
00:05:26,760 --> 00:05:27,440
RECEPTOR-BINDING DOMAIN.

168
00:05:27,440 --> 00:05:28,960
THESE PROTEINS ARE VERY SMALL.

169
00:05:28,960 --> 00:05:30,600
THESE ARE THREE COPIES OF THIS

170
00:05:30,600 --> 00:05:32,720
SMALL PROTEIN SO ONLY 55 AMINO

171
00:05:32,720 --> 00:05:33,640
ACIDS.

172
00:05:33,640 --> 00:05:34,640
IT'S VERY STABLE, IT DOESN'T

173
00:05:34,640 --> 00:05:35,000
MELT.

174
00:05:35,000 --> 00:05:36,600
AND THE STRUCTURES -- SO WE

175
00:05:36,600 --> 00:05:38,800
DESIGN THESE ON THE COMPUTER, WE

176
00:05:38,800 --> 00:05:41,080
DESIGNED A SEQUENCE THAT WAS

177
00:05:41,080 --> 00:05:43,720
JUST FOLDING UP INTO REGULAR

178
00:05:43,720 --> 00:05:44,480
STRUCTURE AND DOCKED TO THE

179
00:05:44,480 --> 00:05:46,120
SPIKE AND IN THIS CRYO-EM

180
00:05:46,120 --> 00:05:48,760
STRUCTURE, YOU CAN SEE THAT THE

181
00:05:48,760 --> 00:05:49,840
DESIGN MODEL IS ACTUALLY VERY

182
00:05:49,840 --> 00:05:53,320
CLOSE TO THE ACTUAL STRUCTURE.

183
00:05:53,320 --> 00:05:55,000
SO WE CAN DESIGN THESE SMALL

184
00:05:55,000 --> 00:05:57,080
PROTEINS VERY ACCURATELY.

185
00:05:57,080 --> 00:05:58,920
NOW, THIS IS A VERY POTENT

186
00:05:58,920 --> 00:06:00,760
PROTEIN, IT PROTECTS MICE AND

187
00:06:00,760 --> 00:06:02,760
HAMSTERS AGAINST THE

188
00:06:02,760 --> 00:06:03,080
CORONAVIRUS.

189
00:06:03,080 --> 00:06:05,480
WHEN ADMINISTERED INTRANASALLY.

190
00:06:05,480 --> 00:06:08,440
TO MAKE IT EVEN MORE POTENT, WE

191
00:06:08,440 --> 00:06:11,280
BASICALLY CONNECTED THREE OF

192
00:06:11,280 --> 00:06:12,840
THESE RECEPTOR LITTLE PROTEINS

193
00:06:12,840 --> 00:06:14,240
THAT BIND TO THE

194
00:06:14,240 --> 00:06:16,520
RECEPTOR-BINDING DOMAIN WITH A

195
00:06:16,520 --> 00:06:17,360
DESIGN TRIMERIC UNIT THAT'S SORT

196
00:06:17,360 --> 00:06:20,040
OF DESIGNED SO THAT IT CAN BIND

197
00:06:20,040 --> 00:06:20,600
ALL THREE AT THE SAME TIME.

198
00:06:20,600 --> 00:06:22,320
HERE'S THE CRYO-EM STRUCTURE,

199
00:06:22,320 --> 00:06:26,520
YOU SEE SORT OF THE TRIPOD-LIKE

200
00:06:26,520 --> 00:06:27,920
STRUCTURE INHIBITOR BOUND TO ALL

201
00:06:27,920 --> 00:06:31,960
TOLETHREE DOMAINS AT THE SAME T.

202
00:06:31,960 --> 00:06:33,720
BECAUSE IT'S BINDING

203
00:06:33,720 --> 00:06:34,600
TRIVALENTLY, IT'S EXTREMELY

204
00:06:34,600 --> 00:06:35,360
TIGHT.

205
00:06:35,360 --> 00:06:36,080
WE CAN'T MEASURE HOW TIGHT IT

206
00:06:36,080 --> 00:06:36,560
IS.

207
00:06:36,560 --> 00:06:37,640
IT NEUTRALIZES ALL OF THE

208
00:06:37,640 --> 00:06:39,520
VARIANTS THAT IT HAS BEEN TESTED

209
00:06:39,520 --> 00:06:44,120
AGAINST WITH VERY HIGH -- AT

210
00:06:44,120 --> 00:06:47,400
VERY HIGH LEVELS, SO AGAIN, IT

211
00:06:47,400 --> 00:06:49,800
CAN GET HARD TO MEASURE

212
00:06:49,800 --> 00:06:53,480
NEUTRALIZATION BELOW 100

213
00:06:53,480 --> 00:06:55,120
PICOMOLAR AND IT ALSO

214
00:06:55,120 --> 00:06:56,680
PROVIDES -- SO BASICALLY THE WAY

215
00:06:56,680 --> 00:06:58,200
IT'S ADMINISTERED IS

216
00:06:58,200 --> 00:06:59,520
INTRANASALLY, YOU CAN GIVE IT UP

217
00:06:59,520 --> 00:07:00,640
TO FOUR DAYS BEFORE EXPOSURE AND

218
00:07:00,640 --> 00:07:01,920
UP TO ONE OR TWO DAYS

219
00:07:01,920 --> 00:07:03,320
AFTERWARDS.

220
00:07:03,320 --> 00:07:06,200
AND IT ALSO BLOCKS TRANSMISSION,

221
00:07:06,200 --> 00:07:08,280
WHICH MAKES SENSE BECAUSE IT'S

222
00:07:08,280 --> 00:07:13,360
BASICALLY JUST BLOCKING THE ACE

223
00:07:13,360 --> 00:07:14,040
2 INTERACTION, BOTH EITHER ON

224
00:07:14,040 --> 00:07:15,800
THE WAY IN OR ON THE WAY OUT.

225
00:07:15,800 --> 00:07:19,400
AND THIS WILL BE STARTING

226
00:07:19,400 --> 00:07:22,600
CLINICAL TRIALS EARLY NEXT YEAR.

227
00:07:22,600 --> 00:07:25,800
SO WE HAVE BEEN EXCITED THEN,

228
00:07:25,800 --> 00:07:27,400
INTERESTED IN SORT OF THIS JUST

229
00:07:27,400 --> 00:07:29,160
IN CASE IDEA, BASICALLY CAN WE

230
00:07:29,160 --> 00:07:31,800
MAKE POTENT INHIBITORS LIKE THIS

231
00:07:31,800 --> 00:07:33,440
FOR OTHER MAJOR CLASSES OF

232
00:07:33,440 --> 00:07:34,960
VIRUSES.

233
00:07:34,960 --> 00:07:36,280
SO -- AND WE'VE BEEN USING THE

234
00:07:36,280 --> 00:07:37,480
SAME BASIC APPROACH WHERE WE

235
00:07:37,480 --> 00:07:38,880
START BY DESIGNING SMALL

236
00:07:38,880 --> 00:07:41,640
PROTEINS, THIS IS THE CASE OF

237
00:07:41,640 --> 00:07:43,160
RSV, WE DESIGNED SMALL PROTEINS

238
00:07:43,160 --> 00:07:45,360
THAT BIND TO THE VIRUS, AND THEN

239
00:07:45,360 --> 00:07:47,400
WE OLIGOMERRIZE THEM, IN THIS

240
00:07:47,400 --> 00:07:55,720
CASE AGAIN, WE HAVE P PICO MOLAR

241
00:07:55,720 --> 00:07:56,840
NEUTRALIZATION OF THESE CHIMERIC

242
00:07:56,840 --> 00:07:57,120
VERSIONS.

243
00:07:57,120 --> 00:08:00,080
THIS IS THE CASE OF MERS, WHERE

244
00:08:00,080 --> 00:08:01,720
AGAIN, WE'RE DOWN WITH

245
00:08:01,720 --> 00:08:07,720
NEUTRALIZATION IC50s IN THE

246
00:08:07,720 --> 00:08:09,040
PICOMOLAR RANGE.

247
00:08:09,040 --> 00:08:10,920
I'M PRETTY CONFIDENT THIS IS A

248
00:08:10,920 --> 00:08:12,360
GENERAL APPROACH FOR MAKING

249
00:08:12,360 --> 00:08:13,320
ANTIVIRALS AND THE NICE THING

250
00:08:13,320 --> 00:08:14,560
ABOUT THESE PROTEINS IS THAT

251
00:08:14,560 --> 00:08:18,160
THEY'RE SO STABLE THAT YOU COULD

252
00:08:18,160 --> 00:08:19,440
COMBINE THEM ALL IN SOME SORT OF

253
00:08:19,440 --> 00:08:21,280
SPRAY THAT INSTEAD OF TAKING

254
00:08:21,280 --> 00:08:22,520
YOUR COVID TEST IN THE MORNING,

255
00:08:22,520 --> 00:08:23,920
YOU WOULD JUST SPRAY A LITTLE

256
00:08:23,920 --> 00:08:25,840
BIT IN YOUR NOSE IF YOU'RE GOING

257
00:08:25,840 --> 00:08:30,360
INTO AN INFECTED AREA.

258
00:08:30,360 --> 00:08:31,880
BUT THERE IS A CAVEAT, WHICH IT

259
00:08:31,880 --> 00:08:33,320
DID TAKE US SOME TIME TO MAKE

260
00:08:33,320 --> 00:08:33,520
THESE.

261
00:08:33,520 --> 00:08:35,480
IT TOOK US -- WE HAD TO SCREEN

262
00:08:35,480 --> 00:08:37,240
QUITE A LARGE NUMBER -- THEY ALL

263
00:08:37,240 --> 00:08:38,720
GET MADE ON COMPUTER BUT THEN WE

264
00:08:38,720 --> 00:08:39,880
HAVE TO TEST A NUMBER.

265
00:08:39,880 --> 00:08:41,520
I'LL SHOW YOU HOW WE CAN MAKE

266
00:08:41,520 --> 00:08:42,720
THESE CLASS OF PROTEINS MUCH

267
00:08:42,720 --> 00:08:43,600
MORE RAPIDLY.

268
00:08:43,600 --> 00:08:45,000
YOU'D REALLY LIKE TO BE ABLE TO

269
00:08:45,000 --> 00:08:47,440
DO THIS IN A NEW VIRUS BREAKS

270
00:08:47,440 --> 00:08:49,520
OUT AND YOU GET THE -- YOU HAVE

271
00:08:49,520 --> 00:08:50,600
THE SEQUENCE AND YOU'D LIKE TO

272
00:08:50,600 --> 00:08:51,960
BE ABLE TO MAKE AN ANTIVIRAL

273
00:08:51,960 --> 00:08:53,400
WITHIN DAYS AND THAT'S OBVIOUSLY

274
00:08:53,400 --> 00:08:55,400
MUCH FASTER THAN YOU CAN DO WITH

275
00:08:55,400 --> 00:08:57,040
ANTIBODIES.

276
00:08:57,040 --> 00:08:58,400
SO WE'VE USED THIS APPROACH TO

277
00:08:58,400 --> 00:09:00,640
MAKE BINDING PROTEINS TO QUITE A

278
00:09:00,640 --> 00:09:03,800
NUMBER OF HUMAN CELL SURFACE

279
00:09:03,800 --> 00:09:05,280
PROTEINS, RECEPTOR SUBUNITS

280
00:09:05,280 --> 00:09:08,000
PRIMARILY.

281
00:09:08,000 --> 00:09:09,960
TYROSINE KINASE SUBRECEPTOR

282
00:09:09,960 --> 00:09:11,200
UNITS, AND WHAT'S EXCITING NOW,

283
00:09:11,200 --> 00:09:13,120
SO WE HAVE ANTAGONISTS FOR ALL

284
00:09:13,120 --> 00:09:14,120
OF THESE, WHICH MAY BE

285
00:09:14,120 --> 00:09:14,920
INTERESTING ON THEIR OWN, BUT

286
00:09:14,920 --> 00:09:16,520
WHAT WE CAN DO IS NOW COMBINE

287
00:09:16,520 --> 00:09:18,520
THEM IN DIFFERENT WAYS TO BRING

288
00:09:18,520 --> 00:09:20,920
RECEPTOR SUBUNITS TOGETHER IN

289
00:09:20,920 --> 00:09:23,200
DIFFERENT ARRANGEMENTS.

290
00:09:23,200 --> 00:09:26,920
AND THAT GIVES A WHOLE HOST OF

291
00:09:26,920 --> 00:09:29,200
NEW AGONIST ACTIVITIES.

292
00:09:29,200 --> 00:09:34,800
SO FOR EXAMPLE, HERE WE HAVE

293
00:09:34,800 --> 00:09:36,840
A -- WE MADE A BINDER AGAINST

294
00:09:36,840 --> 00:09:40,480
THE FGF RECEPTOR, AND WHEN WE

295
00:09:40,480 --> 00:09:43,480
CAN PLACE THAT ON DESIGN

296
00:09:43,480 --> 00:09:46,840
SCAFFOLDS THAT SPACE THESE FGF

297
00:09:46,840 --> 00:09:48,800
BINDING DOMAINS IN DIFFERENT

298
00:09:48,800 --> 00:09:50,560
ARRANGEMENTS AND WE FIND THESE

299
00:09:50,560 --> 00:09:51,640
GIVE DIFFERENT EFFECTS ON CELLS

300
00:09:51,640 --> 00:09:52,600
WHEN THEY'RE ADDED.

301
00:09:52,600 --> 00:09:54,720
SO HERE YOU SEE IN THIS

302
00:09:54,720 --> 00:09:56,160
ARRANGEMENT, WE GET POTENT

303
00:09:56,160 --> 00:09:57,440
CALCIUM SIGNALING BUT IN THIS

304
00:09:57,440 --> 00:09:59,080
ARRANGEMENT, WE DON'T GET ANY --

305
00:09:59,080 --> 00:10:00,280
WE GET MUCH LESS SIGNALING.

306
00:10:00,280 --> 00:10:02,440
SO WE CAN TUNE THE RESPONSE BY

307
00:10:02,440 --> 00:10:05,440
CHANGING THE ARRANGEMENT.

308
00:10:05,440 --> 00:10:08,680
AND THIS IS -- THESE COMPOUNDS

309
00:10:08,680 --> 00:10:11,000
HAVE VERY INTERESTING EFFECTS.

310
00:10:11,000 --> 00:10:13,080
THIS IS -- THAT DIFFER FROM FGF,

311
00:10:13,080 --> 00:10:14,920
SO THIS IS LOOKING AT VASCULAR

312
00:10:14,920 --> 00:10:15,880
DIFFERENTIATION, AND YOU CAN SEE

313
00:10:15,880 --> 00:10:19,440
THE RATIO DURING ENDOTHELIAL

314
00:10:19,440 --> 00:10:20,720
CELL DIFFERENTIATION, THE RATIO

315
00:10:20,720 --> 00:10:23,240
OF ARTERIAL TO VENOUS CELLS THAT

316
00:10:23,240 --> 00:10:26,000
ARE PRODUCED DIFFERS

317
00:10:26,000 --> 00:10:27,400
DRAMATICALLY BETWEEN FGF, WHICH

318
00:10:27,400 --> 00:10:29,040
IS PRIMARILY GOING DOWN THE

319
00:10:29,040 --> 00:10:30,920
VENOUS PATHWAY, WHEREAS THE

320
00:10:30,920 --> 00:10:32,640
SYNTHETIC AGONIST DOWN HERE IS

321
00:10:32,640 --> 00:10:33,840
PRIMARILY GOING DOWN THE

322
00:10:33,840 --> 00:10:36,880
ARTERIAL PATHWAY.

323
00:10:36,880 --> 00:10:38,120
THIS IS, I THINK, REALLY JUST

324
00:10:38,120 --> 00:10:39,240
THE TIP OF THE ICEBERG.

325
00:10:39,240 --> 00:10:40,760
THIS IS AN EXAMPLE HERE WHERE WE

326
00:10:40,760 --> 00:10:42,320
CAN TAKE TWO DIFFERENT BINDERS,

327
00:10:42,320 --> 00:10:45,360
ONE AGAINST THE IL-7 RECEPTOR,

328
00:10:45,360 --> 00:10:46,360
ONE AGAINST THE COMMON GAMMA

329
00:10:46,360 --> 00:10:49,720
CHAIN AND MAKE EXTREMELY POTENT

330
00:10:49,720 --> 00:10:50,240
IL-7 MIMICS.

331
00:10:50,240 --> 00:10:51,800
SO NOW WE'RE EXPLORING DIFFERENT

332
00:10:51,800 --> 00:10:54,200
COMBINATIONS OF THESE LITTLE

333
00:10:54,200 --> 00:10:55,840
DOMAINS, AND SORT OF SEEING WHAT

334
00:10:55,840 --> 00:10:59,280
THE RICHNESS OF INDUCED EFFECTS

335
00:10:59,280 --> 00:10:59,560
CAN BE.

336
00:10:59,560 --> 00:11:02,400
SO I THINK THERE'S A LOT OF

337
00:11:02,400 --> 00:11:04,800
OPPORTUNITY FOR REGENERATIVE

338
00:11:04,800 --> 00:11:06,200
MEDICINE, YOU COULD EVEN IMAGINE

339
00:11:06,200 --> 00:11:08,040
THINGS LIKE TRANS

340
00:11:08,040 --> 00:11:10,800
DIFFERENTIATION, AND OTHER --

341
00:11:10,800 --> 00:11:16,560
AND IN THE CASE OF THE IMMUNE

342
00:11:16,560 --> 00:11:19,400
SYSTEM, BEING ABLE TO -- IT'S

343
00:11:19,400 --> 00:11:25,320
KIND OF AN EXCITING TIME.

344
00:11:25,320 --> 00:11:28,600
SO I TALKED SO FAR ABOUT MAKING

345
00:11:28,600 --> 00:11:29,560
BINDERS AGAINST FOLDED

346
00:11:29,560 --> 00:11:30,680
STRUCTURES, BUT WE CAN ALSO NOW

347
00:11:30,680 --> 00:11:33,160
MAKE BINDERS AGAINST STRUCTURES

348
00:11:33,160 --> 00:11:34,520
WHICH AREN'T FOLDED.

349
00:11:34,520 --> 00:11:38,720
SO IN PARTICULAR, WE CAN MAKE

350
00:11:38,720 --> 00:11:40,080
BINDERS TO PEPTIDES WHICH FORM

351
00:11:40,080 --> 00:11:40,400
AMYLOID.

352
00:11:40,400 --> 00:11:44,760
THIS IS TO THE -- TO A BETA, AND

353
00:11:44,760 --> 00:11:46,280
THE BASIC IDEA IS WE CAN BIND IT

354
00:11:46,280 --> 00:11:50,680
WHEN IT'S IN A BETA STRAND

355
00:11:50,680 --> 00:11:51,680
CONFIRMATION, AND HERE YOU SEE

356
00:11:51,680 --> 00:11:52,760
THE BASIC IDEA, HERE'S THE

357
00:11:52,760 --> 00:11:55,480
DESIGN AND THERE'S THE PEPTIDE

358
00:11:55,480 --> 00:11:56,120
SLOTTED IN.

359
00:11:56,120 --> 00:11:58,520
AND THESE PROTEINS BIND TO THEIR

360
00:11:58,520 --> 00:12:02,600
TARGET PEPTIDES QUITE TIGHTLY,

361
00:12:02,600 --> 00:12:05,120
AND OF COURSE A BETA, IF YOU

362
00:12:05,120 --> 00:12:06,640
JUST LET IT FOLD ON ITS OWN,

363
00:12:06,640 --> 00:12:09,800
WILL ASSEMBLE INTO THESE AMYLOID

364
00:12:09,800 --> 00:12:10,240
FIBRILS.

365
00:12:10,240 --> 00:12:11,800
IF THESE DESIGN PROTEINS ARE

366
00:12:11,800 --> 00:12:13,240
ADDED, THEY COMPLETELY SHUT THIS

367
00:12:13,240 --> 00:12:14,720
PROCESS DOWN AND SO THERE'S A

368
00:12:14,720 --> 00:12:20,440
COLLABORATION WITH THOMAS

369
00:12:20,440 --> 00:12:23,600
NKOWLES LAB, AND HERE A BETA

370
00:12:23,600 --> 00:12:25,240
AGGREGATE VERY RAPIDLY BUT WHEN

371
00:12:25,240 --> 00:12:30,480
THE PROTEINS ARE ADDED, THERE'S

372
00:12:30,480 --> 00:12:33,120
A COMPLETE SUPPRESSION OF

373
00:12:33,120 --> 00:12:34,720
AGGREGATION OVER MULTIPLE DAYS.

374
00:12:34,720 --> 00:12:36,360
AND THIS IS ACTUALLY MUCH

375
00:12:36,360 --> 00:12:37,120
STRONGER THAN HAS BEEN OBSERVED

376
00:12:37,120 --> 00:12:41,320
FOR THE THERAPEUTIC ANTIBODIES.

377
00:12:41,320 --> 00:12:44,480
SO THE ROLE OF THESE

378
00:12:44,480 --> 00:12:47,120
AMYLOID-LIKE STRUCTURES IS STILL

379
00:12:47,120 --> 00:12:48,320
UNCLEAR, BUT WE THINK THIS CLASS

380
00:12:48,320 --> 00:12:49,520
OF PROTEINS WILL PROVIDE A

381
00:12:49,520 --> 00:12:51,320
CLEANER WAY TO PREVENT

382
00:12:51,320 --> 00:12:54,080
AGGREGATION AND SHOULD ALSO

383
00:12:54,080 --> 00:12:57,480
SERVE AS DIAGNOSTICS.

384
00:12:57,480 --> 00:13:00,240
SO ON THIS THEME OF BINDING TO

385
00:13:00,240 --> 00:13:03,000
DISORDERED PARTS OF PROTEINS,

386
00:13:03,000 --> 00:13:04,520
THESE HAVE BEEN HARD TO TARGET

387
00:13:04,520 --> 00:13:09,120
BY CONVENTIONAL METHODS.

388
00:13:09,120 --> 00:13:10,520
PROTEINS BIND TO EXTENDED

389
00:13:10,520 --> 00:13:12,480
PEPTIDES SUCH THAT EVERY OTHER

390
00:13:12,480 --> 00:13:13,600
AMINO ACID OF THE PEPTIDE FITS

391
00:13:13,600 --> 00:13:16,560
INTO A POCKET ON THE PROTEIN.

392
00:13:16,560 --> 00:13:18,600
AND SHE CAN CHANGE THE POCKETS

393
00:13:18,600 --> 00:13:21,000
HERE TO GET SPECIFICITIES FOR

394
00:13:21,000 --> 00:13:22,400
DIFFERENT AMINO ACID SEQUENCES.

395
00:13:22,400 --> 00:13:24,720
AND THESE DESIGN PROTEINS, THE

396
00:13:24,720 --> 00:13:26,760
SEQUENCES THEY BIND TO, ARE, FOR

397
00:13:26,760 --> 00:13:27,880
EXAMPLE, SHOWN HERE.

398
00:13:27,880 --> 00:13:30,040
THIS IS THE SAME POCKET REPEATED

399
00:13:30,040 --> 00:13:33,160
SIX TIMES, SO IT'S THE SEQUENCE

400
00:13:33,160 --> 00:13:35,440
LEUCINE, ARGININE, PROLINE

401
00:13:35,440 --> 00:13:35,880
REPEATED SIX TIMES.

402
00:13:35,880 --> 00:13:37,640
AS YOU CAN SEE HERE FROM THESE

403
00:13:37,640 --> 00:13:41,240
KDs, THEY BIND VERY TIGHTLY,

404
00:13:41,240 --> 00:13:43,960
AND THEY BIND BOTH IN VITRO AND

405
00:13:43,960 --> 00:13:48,560
IN CELLS TO THEIR TARGETS.

406
00:13:48,560 --> 00:13:51,200
AND WHAT'S NEAT IS, NOW SHE CAN

407
00:13:51,200 --> 00:13:55,600
TARGET INTRINSICALLY DISORDERED

408
00:13:55,600 --> 00:13:57,000
REGIONS OF HUMAN PROTEINS, SO

409
00:13:57,000 --> 00:13:58,760
SHE CAN PICK OUT REGIONS WITH --

410
00:13:58,760 --> 00:14:01,040
AND THEN BASICALLY MAKE MODULAR

411
00:14:01,040 --> 00:14:05,640
PROTEINS THAT BIND TO THAT

412
00:14:05,640 --> 00:14:07,280
PEPTIDE SEQUENCE AND SHE CAN USE

413
00:14:07,280 --> 00:14:08,280
THESE PROTEINS, THIS IS ACTUALLY

414
00:14:08,280 --> 00:14:17,560
A COLLABORATIO COLLABORATION WIL

415
00:14:17,560 --> 00:14:18,880
DERIVERY, AND SEE WHAT PROTEINS

416
00:14:18,880 --> 00:14:19,440
COME WITH THEM.

417
00:14:19,440 --> 00:14:20,600
SO THAT'S PRETTY MUCH WHERE WE

418
00:14:20,600 --> 00:14:22,240
ARE WITH USING SORT OF THIS

419
00:14:22,240 --> 00:14:23,320
ENERGY-BASED APPROACH TO

420
00:14:23,320 --> 00:14:24,800
DESIGNING BINDERS, AND AGAIN,

421
00:14:24,800 --> 00:14:25,880
I'LL COME BACK AT THE END OF MY

422
00:14:25,880 --> 00:14:27,600
TALK TO YOU DOING THIS WITH DEEP

423
00:14:27,600 --> 00:14:30,600
LEARNING.

424
00:14:30,600 --> 00:14:31,440
WE'VE ALSO BEEN INTERESTED IN

425
00:14:31,440 --> 00:14:32,880
THE PROBLEM OF DELIVERING

426
00:14:32,880 --> 00:14:34,200
BIOLOGICS, SO HOW DO YOU DELIVER

427
00:14:34,200 --> 00:14:36,680
GUIDE RNAs OR PROTEIN

428
00:14:36,680 --> 00:14:40,960
THERAPEUTICS INTO CELLS?

429
00:14:40,960 --> 00:14:42,720
WE DESIGNED THESE ANTIBODY

430
00:14:42,720 --> 00:14:43,480
NANOCAGES.

431
00:14:43,480 --> 00:14:45,000
THEY'RE MADE OUT OF A DESIGN

432
00:14:45,000 --> 00:14:46,240
PROTEIN LIKE THIS PENT MER IN

433
00:14:46,240 --> 00:14:48,920
GREEN HERE AND THEY BIND TO

434
00:14:48,920 --> 00:14:49,920
ANTIBODIES AS SHOWN HERE, AND

435
00:14:49,920 --> 00:14:54,960
THEY BASICALLY WRAP THE ANTIBODY

436
00:14:54,960 --> 00:14:55,960
INTO A NANOCAGE AND THAT'S SHOWN

437
00:14:55,960 --> 00:14:56,280
HERE.

438
00:14:56,280 --> 00:14:59,600
WE HAVE OTHER ONES THAT WILL

439
00:14:59,600 --> 00:15:05,840
WRAP ANTIBODIES INTO OCTAHEDRAL

440
00:15:05,840 --> 00:15:06,440
CAGES.

441
00:15:06,440 --> 00:15:07,440
THIS IS NICE BECAUSE YOU CAN

442
00:15:07,440 --> 00:15:10,280
TAKE ANY ANTIBODY, ADD THE

443
00:15:10,280 --> 00:15:11,600
CORRESPONDING PROTEIN AND YOU

444
00:15:11,600 --> 00:15:13,800
WILL GET A NICE FOLDED SORT OF

445
00:15:13,800 --> 00:15:17,920
VIRUS-LIKE CAGE FROM IT.

446
00:15:17,920 --> 00:15:20,920
SO FROM THE DELIVERY POINT OF

447
00:15:20,920 --> 00:15:22,480
VIEW, WE CAN BASICALLY PUT ON

448
00:15:22,480 --> 00:15:24,360
THE CAGES AN ANTIBODY WHICH

449
00:15:24,360 --> 00:15:27,680
TARGETS THEM TO A PARTICULAR --

450
00:15:27,680 --> 00:15:30,840
TO PARTICULAR CELLS IN THE BODY.

451
00:15:30,840 --> 00:15:33,240
BUT THEN THE PARTICLES GET TAKEN

452
00:15:33,240 --> 00:15:34,800
UP WITH WHATEVER WE PUT INSIDE

453
00:15:34,800 --> 00:15:36,240
INTO CELLS BUT THEN THEY HAVE TO

454
00:15:36,240 --> 00:15:41,040
GET OUT OF THE ENDO SOME.

455
00:15:41,040 --> 00:15:42,280
WHAT AARON YOUNG HAS DONE IS

456
00:15:42,280 --> 00:15:48,920
FIRST OF ALL TO KEEP THE CARGO

457
00:15:48,920 --> 00:15:50,560
IN, SO IF YOU LOOK AT THIS

458
00:15:50,560 --> 00:15:51,440
ORIGINAL ONE HERE, THERE'S A

459
00:15:51,440 --> 00:15:53,400
HOLE HERE.

460
00:15:53,400 --> 00:15:55,600
AND WHAT ERIN YOUNG DID, A

461
00:15:55,600 --> 00:15:56,800
GRADUATE STUDENT IN THE LAB, SHE

462
00:15:56,800 --> 00:15:58,000
DESIGNED A TRIMER THAT FITS

463
00:15:58,000 --> 00:16:00,920
RIGHT IN THERE.

464
00:16:00,920 --> 00:16:02,880
HERE'S THE CRYO-EM STRUCTURE

465
00:16:02,880 --> 00:16:04,600
NOW, AND YOU CAN SEE DOWN THE

466
00:16:04,600 --> 00:16:09,400
THREE FOLD AND FOUR FOLD AXES OF

467
00:16:09,400 --> 00:16:10,480
THE DESIGN, HERE'S WHERE THE

468
00:16:10,480 --> 00:16:12,680
PLUG SITS, HERE'S WHERE THE

469
00:16:12,680 --> 00:16:13,360
ANTIBODY SITS.

470
00:16:13,360 --> 00:16:15,840
SHE'S DESIGNED THIS TO BE

471
00:16:15,840 --> 00:16:16,960
PH-DEPENDENT SO WHEN IT'S TAKEN

472
00:16:16,960 --> 00:16:18,800
UP INTO THE ENDOSOME THE PLUG

473
00:16:18,800 --> 00:16:20,920
COMES OUT AND WE'RE NOW TRYING

474
00:16:20,920 --> 00:16:22,520
TO OPTIMIZE THE ENTO SOMOL

475
00:16:22,520 --> 00:16:24,080
ESCAPE SO WHAT'S INSIDE THE CAGE

476
00:16:24,080 --> 00:16:25,720
CAN NOW COME OUT BECAUSE IT GETS

477
00:16:25,720 --> 00:16:26,800
OUT OF THE CAGE WHEN THE PLUG

478
00:16:26,800 --> 00:16:28,200
COMES OUT BUT IT'S STILL GOT TO

479
00:16:28,200 --> 00:16:30,160
GET ACROSS THE ENDO SOMOL

480
00:16:30,160 --> 00:16:32,560
MEMBRANE.

481
00:16:32,560 --> 00:16:34,440
SO I'VE TALKED SO FAR ABOUT

482
00:16:34,440 --> 00:16:36,960
SOLUBLE PROTEINS BUT WE CAN ALSO

483
00:16:36,960 --> 00:16:39,160
NOW DESIGN TRANSMEMBRANE

484
00:16:39,160 --> 00:16:39,760
PROTEINS.

485
00:16:39,760 --> 00:16:47,160
SO HERE IS A DESIGNED TRANS

486
00:16:47,160 --> 00:16:48,800
MEMBRANE BETA BARREL.

487
00:16:48,800 --> 00:16:51,240
HERE IS THE CRYSTAL STRUCTURE OF

488
00:16:51,240 --> 00:16:53,840
IT, YOU CAN SEE THEY'RE CLOSE.

489
00:16:53,840 --> 00:16:55,040
SINCE WE'RE DESIGNING THEM, WE

490
00:16:55,040 --> 00:16:56,240
CAN CHANGE THE INTERIOR TO BE

491
00:16:56,240 --> 00:17:00,640
ANYTHING WE WANT, AND THEN GET

492
00:17:00,640 --> 00:17:01,720
SELECTIVITY FOR DIFFERENT TYPES

493
00:17:01,720 --> 00:17:02,520
OF COMPOUNDS.

494
00:17:02,520 --> 00:17:06,640
AS MANY OF YOU KNOW, SINGLE

495
00:17:06,640 --> 00:17:08,160
MOLECULE DNA SEQUENCING CAN BE

496
00:17:08,160 --> 00:17:09,800
DONE WITH NATURALLY OCCURRING

497
00:17:09,800 --> 00:17:11,360
PORES BUT WE CAN REALLY OPTIMIZE

498
00:17:11,360 --> 00:17:13,120
THEM FOR THAT OR FOR PEPTIDE

499
00:17:13,120 --> 00:17:13,800
SEQUENCING OR FOR SOMETHING

500
00:17:13,800 --> 00:17:14,520
ELSE.

501
00:17:14,520 --> 00:17:17,360
AND THESE DESIGN PORES HAVE --

502
00:17:17,360 --> 00:17:18,800
ARE VERY STABLE, THEY HAVE THESE

503
00:17:18,800 --> 00:17:23,480
VERY CLEAR SINGLE CHAJ

504
00:17:23,480 --> 00:17:24,560
CONDUCTANCES AS YOU SEE HERE, SO

505
00:17:24,560 --> 00:17:25,880
WE'RE NOW DESIGNING PROTEINS

506
00:17:25,880 --> 00:17:29,400
WITH DIFFERENT SIZE AND SHAPE

507
00:17:29,400 --> 00:17:30,480
CHANNELS.

508
00:17:30,480 --> 00:17:32,320
SO ONE OF THE THINGS THAT IF YOU

509
00:17:32,320 --> 00:17:33,520
HAVEN'T LOOKED AT A LOT OF

510
00:17:33,520 --> 00:17:37,080
PROTEINS BEFORE, YOU PROBABLY

511
00:17:37,080 --> 00:17:37,960
FIND THESE PICTURES KIND OF HARD

512
00:17:37,960 --> 00:17:38,840
TO LOOK AT.

513
00:17:38,840 --> 00:17:40,160
GENERALLY WHEN THE FIRST PROTEIN

514
00:17:40,160 --> 00:17:41,200
STRUCTURES WERE SOLVED, IT WAS

515
00:17:41,200 --> 00:17:43,640
FELT THEY'RE SO COMPLICATED AND

516
00:17:43,640 --> 00:17:44,520
ASYMMETRIC, AND THAT'S BECAUSE

517
00:17:44,520 --> 00:17:46,360
NATURE REALLY NEVER CARED ABOUT

518
00:17:46,360 --> 00:17:50,360
CREATING PROTEINS THAT WERE EASY

519
00:17:50,360 --> 00:17:52,640
FOR PEOPLE TO LOOK AT.

520
00:17:52,640 --> 00:17:54,920
BUT TIM HUDDY, GRAD STUDENT IN

521
00:17:54,920 --> 00:17:56,160
THE GROUP, IT DIDN'T HAVE TO BE

522
00:17:56,160 --> 00:17:58,080
THAT CASE WAY, SO HE'S DESIGNED

523
00:17:58,080 --> 00:17:59,880
PROTEINS WHICH ARE REALLY MORE

524
00:17:59,880 --> 00:18:00,840
LIKE THE BUILDING BLOCKS YOU

525
00:18:00,840 --> 00:18:02,280
WOULD USE TO, SAY, CONSTRUCT A

526
00:18:02,280 --> 00:18:04,320
HOUSE.

527
00:18:04,320 --> 00:18:06,320
SO HE'S BUILT THESE TOTALLY

528
00:18:06,320 --> 00:18:07,320
STRAIGHT PROTEINS, WHICH ARE

529
00:18:07,320 --> 00:18:10,080
LIKE THESE KIND OF BLOCKS HERE,

530
00:18:10,080 --> 00:18:11,280
AND SINCE THEY'RE JUST MADE OUT

531
00:18:11,280 --> 00:18:12,800
OF REPEATING ELEMENTS, HE CAN

532
00:18:12,800 --> 00:18:15,320
MAKE THEM LONGER, HE CAN PUT IN

533
00:18:15,320 --> 00:18:18,720
RIGHT ANGLE TURNS LIKE THIS, HE

534
00:18:18,720 --> 00:18:21,440
CAN MAKE CURVED VERSIONS LIKE

535
00:18:21,440 --> 00:18:23,280
THIS AND THEN MAKE ONES THAT

536
00:18:23,280 --> 00:18:24,480
INTERACT -- PUT IN INTERFACES

537
00:18:24,480 --> 00:18:26,000
BETWEEN THE BLOCK SO THEY CAN

538
00:18:26,000 --> 00:18:26,880
INTERACT IN DIFFERENT WAYS.

539
00:18:26,880 --> 00:18:28,440
AND FROM THESE, HE CAN BUILD ALL

540
00:18:28,440 --> 00:18:30,400
KINDS OF COOL THINGS.

541
00:18:30,400 --> 00:18:32,920
THIS IS JUST -- UP HERE IS A

542
00:18:32,920 --> 00:18:34,560
CRYSTAL STRUCTURE OF ONE OF THE

543
00:18:34,560 --> 00:18:35,120
BLOCKS.

544
00:18:35,120 --> 00:18:36,520
HERE'S SHOWING WITH DIFFERENT

545
00:18:36,520 --> 00:18:38,480
TYPES OF CORNERS, HE CAN MAKE

546
00:18:38,480 --> 00:18:39,920
TRIANGLES, THIS IS ELECTRON

547
00:18:39,920 --> 00:18:43,120
DENSITY DATA, AND -- OR SQUARES

548
00:18:43,120 --> 00:18:44,520
AND BY PUTTING THE CURVE BLOCKS

549
00:18:44,520 --> 00:18:46,280
TOGETHER, HE CAN MAKE THESE VERY

550
00:18:46,280 --> 00:18:47,360
HOMOGENEOUS POPULATIONS OF

551
00:18:47,360 --> 00:18:51,720
CIRCLES.

552
00:18:51,720 --> 00:18:53,160
NOW BY COMBINING THE CURVED

553
00:18:53,160 --> 00:18:56,400
BLOCKS HERE WITH THESE SORT OF

554
00:18:56,400 --> 00:18:58,000
INTERACTION BLOCKS, HE CAN BUILD

555
00:18:58,000 --> 00:19:01,600
DIFFERENT TYPES OF POLYHEDRA.

556
00:19:01,600 --> 00:19:05,280
THESE ARE EM RECONSTRUCTIONS

557
00:19:05,280 --> 00:19:06,480
REALLY BY JUST PUTTING BLOCKS

558
00:19:06,480 --> 00:19:10,160
TOGETHER, AND YOU CAN SEE THAT

559
00:19:10,160 --> 00:19:12,040
THE ELECTRON DENSITY GOES REALLY

560
00:19:12,040 --> 00:19:13,760
WELL WITH THE DESIGN MODEL.

561
00:19:13,760 --> 00:19:15,520
THE EASIEST ONE TO SEE IS MAYBE

562
00:19:15,520 --> 00:19:16,720
THIS ONE HERE, MADE BY SORT OF

563
00:19:16,720 --> 00:19:18,560
TAKING THIS BLOCK WITH FOUR

564
00:19:18,560 --> 00:19:19,560
SUBUNITS, KIND OF LIKE IN A

565
00:19:19,560 --> 00:19:22,840
CROSS SHAPE, PUTTING IT TOGETHER

566
00:19:22,840 --> 00:19:24,600
WITH THIS RIGHT ANGLE

567
00:19:24,600 --> 00:19:26,000
INTERACTION AND WHAT THAT GIVES

568
00:19:26,000 --> 00:19:27,000
IS THIS CUBE SHOWN HERE.

569
00:19:27,000 --> 00:19:28,400
AND SINCE THESE BLOCKS ARE

570
00:19:28,400 --> 00:19:31,920
PERFECTLY STRAIGHT, HE CAN

571
00:19:31,920 --> 00:19:32,880
SIMPLY CHANGE THE LENGTH OF THIS

572
00:19:32,880 --> 00:19:35,080
BLOCK TO MAKE EITHER A SHORT

573
00:19:35,080 --> 00:19:39,680
CUBE, MEDIUM SIZED CUBE OR A BIG

574
00:19:39,680 --> 00:19:40,840
CUBE, AND ALL THAT'S HAPPENING

575
00:19:40,840 --> 00:19:42,080
IF YOU LOOK AT THESE PICTURES IS

576
00:19:42,080 --> 00:19:43,160
THAT THIS IS GETTING LONGER.

577
00:19:43,160 --> 00:19:44,720
THIS IS, AGAIN, ELECTRON

578
00:19:44,720 --> 00:19:45,720
DENSITY.

579
00:19:45,720 --> 00:19:52,280
SO HE CAN ALSO MAKE UNBOUNDED

580
00:19:52,280 --> 00:19:53,360
STRUCTURES LIKE THESE, WHICH HE

581
00:19:53,360 --> 00:19:54,880
CALLS TRAIN TRACKS, WHICH IS

582
00:19:54,880 --> 00:19:55,960
PUTTING TOGETHER THESE BLOCKS

583
00:19:55,960 --> 00:19:58,400
AND THESE BLOCKS, WHICH -- SO HE

584
00:19:58,400 --> 00:20:00,000
GETS THESE TWO PARALLEL THINGS

585
00:20:00,000 --> 00:20:02,000
WITH THIS CONNECTOR HERE, AND

586
00:20:02,000 --> 00:20:03,520
YOU CAN SEE PICTURES OF THEM

587
00:20:03,520 --> 00:20:04,640
HERE.

588
00:20:04,640 --> 00:20:06,400
AND THEN FINALLY, I SHOWED YOU

589
00:20:06,400 --> 00:20:07,440
THOSE RINGS BEFORE.

590
00:20:07,440 --> 00:20:10,280
HE CAN TAKE AN INNER RING AND AN

591
00:20:10,280 --> 00:20:11,520
OUTER RING AND COMBINE IT WITH

592
00:20:11,520 --> 00:20:12,840
THESE BLOCKS TO MAKE THESE SORTS

593
00:20:12,840 --> 00:20:14,480
OF STRUCTURES HERE.

594
00:20:14,480 --> 00:20:16,080
WHICH WE'RE VERY EXCITED ABOUT

595
00:20:16,080 --> 00:20:19,160
AS LARGER PORES THAT COULD FIT

596
00:20:19,160 --> 00:20:23,720
INTO SILICON WAFERS TO ENABLE

597
00:20:23,720 --> 00:20:24,960
COUPLING WITH WHATEVER'S GOING

598
00:20:24,960 --> 00:20:26,520
THROUGH THE ELECTRONIC READOUT.

599
00:20:26,520 --> 00:20:28,000
SO SO FAR, EVERYTHING I'VE TOLD

600
00:20:28,000 --> 00:20:29,240
YOU ABOUT IS DESIGNING PROTEINS

601
00:20:29,240 --> 00:20:31,200
THAT ARE VERY RIGID, THAT JUST

602
00:20:31,200 --> 00:20:35,160
HAVE ONE STRUCTURE.

603
00:20:35,160 --> 00:20:37,560
AND SO THEY WERE INTERESTED IN

604
00:20:37,560 --> 00:20:39,040
DESIGNING PROTEINS THAT HAVE TWO

605
00:20:39,040 --> 00:20:40,920
DIFFERENT STRUCTURES, BUT VERY

606
00:20:40,920 --> 00:20:41,480
DISCRETE.

607
00:20:41,480 --> 00:20:43,120
SO HERE THIS IS THE SYSTEM, SO

608
00:20:43,120 --> 00:20:45,160
THEY DESIGNED -- THERE'S ONE

609
00:20:45,160 --> 00:20:46,480
STATE HERE AND ANOTHER STATE

610
00:20:46,480 --> 00:20:48,360
WHERE THIS BLUE SORT OF ROTATE

611
00:20:48,360 --> 00:20:49,880
AROUND THE ORANGE.

612
00:20:49,880 --> 00:20:52,120
AND ONLY IN THIS STATE WILL IT

613
00:20:52,120 --> 00:20:58,440
BIND AN EFFECTOR, A PEPTIDE.

614
00:20:58,440 --> 00:21:04,000
SO HERE, THIS IS DEER

615
00:21:04,000 --> 00:21:05,200
SPECTROSCOPY WHICH MONITORING

616
00:21:05,200 --> 00:21:07,200
HOW FAR APART PROBE SPIN LABELS

617
00:21:07,200 --> 00:21:08,400
THAT HAVE BEEN PLACED ON THE

618
00:21:08,400 --> 00:21:09,040
PROTEIN ARE.

619
00:21:09,040 --> 00:21:11,240
SO IN THE PRESENCE OF PEPTIDE,

620
00:21:11,240 --> 00:21:12,640
THESE TWO PROBES IN THE MODEL

621
00:21:12,640 --> 00:21:18,360
COME CLOSER AND INDEED, THIS IS

622
00:21:18,360 --> 00:21:20,080
THE DISTANCE AS READ OUT USING

623
00:21:20,080 --> 00:21:21,080
THIS DEER MESH.

624
00:21:21,080 --> 00:21:24,360
WMEASUREMENT.WHEN THE PEPTIDE IT

625
00:21:24,360 --> 00:21:26,600
THE PROBES ARE FARTHER APART AND

626
00:21:26,600 --> 00:21:28,920
YOU SEE THE DISTANCES MOVE TO

627
00:21:28,920 --> 00:21:29,920
FURTHER APART.

628
00:21:29,920 --> 00:21:31,800
SO WHEN THE PEPTIDE IS ADDED,

629
00:21:31,800 --> 00:21:36,800
YOU SWITCH FROM ONE STATE TO

630
00:21:36,800 --> 00:21:37,160
ANOTHER.

631
00:21:37,160 --> 00:21:38,480
THEY WERE ABLE TO SOLVE THE

632
00:21:38,480 --> 00:21:39,440
CRYSTAL STRUCTURE OF THIS IN

633
00:21:39,440 --> 00:21:39,880
BOTH STATES.

634
00:21:39,880 --> 00:21:41,240
SO HERE'S THE CRYSTAL STRUCTURE

635
00:21:41,240 --> 00:21:43,400
IN THE ABSENCE OF THE PEPTIDE.

636
00:21:43,400 --> 00:21:44,920
IT'S BASICALLY IDENTICAL TO THE

637
00:21:44,920 --> 00:21:46,440
DESIGN MODEL, AND HERE'S THE

638
00:21:46,440 --> 00:21:47,240
CRYSTAL STRUCTURE WITH THE

639
00:21:47,240 --> 00:21:48,320
PEPTIDE, AND YOU CAN SEE IT'S

640
00:21:48,320 --> 00:21:51,280
IDENTICAL TO THE SECOND STATE.

641
00:21:51,280 --> 00:21:53,240
SO AGAIN, WE HAVE TWO VERY

642
00:21:53,240 --> 00:21:55,760
WELL-DEFINED STATES.

643
00:21:55,760 --> 00:21:58,120
SO THEY'VE TAKEN THE SYSTEM AND

644
00:21:58,120 --> 00:21:59,560
BUILT ALLOSTERIC ASSEMBLIES.

645
00:21:59,560 --> 00:22:00,320
HERE'S THE BASIC IDEA.

646
00:22:00,320 --> 00:22:05,160
SO THE SYSTEM IS LIKE THIS

647
00:22:05,160 --> 00:22:06,920
STRAIGHT PROTEIN, WHICH WHEN YOU

648
00:22:06,920 --> 00:22:09,840
ADD THE EFFECTOR, IT BENDS.

649
00:22:09,840 --> 00:22:11,400
THIS IS KIND OF AN IDEALIZED

650
00:22:11,400 --> 00:22:11,800
CASE.

651
00:22:11,800 --> 00:22:13,880
SO IF THESE UNITS HAVE AN

652
00:22:13,880 --> 00:22:15,120
INTERFACE ON THE END THAT ALLOWS

653
00:22:15,120 --> 00:22:18,040
THEM TO INTERACT LIKE THIS, THEN

654
00:22:18,040 --> 00:22:19,840
WHEN THE EFFECTOR IS ADDED, THEN

655
00:22:19,840 --> 00:22:21,320
THAT WILL CHANGE THE OUTCOME

656
00:22:21,320 --> 00:22:25,720
PAUSE NOW THE GEOBECAUSE NOW THF

657
00:22:25,720 --> 00:22:26,600
THESE ARE DIFFERENT, SO HE'S

658
00:22:26,600 --> 00:22:28,640
BEEN ABLE TO MAKE A SERIES OF

659
00:22:28,640 --> 00:22:29,080
DIFFERENT ASSEMBLIES.

660
00:22:29,080 --> 00:22:30,520
SO THIS ONE INVOLVES TWO OF

661
00:22:30,520 --> 00:22:31,840
THESE COMING TOGETHER IN THE

662
00:22:31,840 --> 00:22:33,120
ABSENCE OF THE PEPTIDE, BUT WHEN

663
00:22:33,120 --> 00:22:35,440
THE PEPTIDE BINDS, THE

664
00:22:35,440 --> 00:22:36,440
CONFORMATION CHANGES AND NOW HE

665
00:22:36,440 --> 00:22:39,840
GETS THESE TRIMERS, THESE

666
00:22:39,840 --> 00:22:40,720
TRIANGULAR STRUCTURES.

667
00:22:40,720 --> 00:22:43,120
HERE'S ONE THAT GOES FROM BEING

668
00:22:43,120 --> 00:22:44,960
A CYCLIC TRIMER WITH THREE

669
00:22:44,960 --> 00:22:50,200
COPIES, TO A TETRAMER, WHEN THE

670
00:22:50,200 --> 00:22:56,880
PEPTIDE IS ADDED.

671
00:22:56,880 --> 00:22:57,880
SORRY, I SKIPPED ONE.

672
00:22:57,880 --> 00:22:59,640
WE'RE ABOUT TO GET TO DEEP

673
00:22:59,640 --> 00:23:01,680
LEARNING, AS YOU CAN SEE.

674
00:23:01,680 --> 00:23:03,360
SO IT'S NEAT, SO WE CAN DESIGN

675
00:23:03,360 --> 00:23:04,680
THESE PROTEINS THAT SWITCH

676
00:23:04,680 --> 00:23:05,440
BETWEEN TWO STATES AND WE CAN

677
00:23:05,440 --> 00:23:08,600
USE THEM TO DERIVE CHANGES IN

678
00:23:08,600 --> 00:23:12,720
ASSEMBLY STATE.

679
00:23:12,720 --> 00:23:13,760
HERE'S A PARTICULARLY NICE ONE

680
00:23:13,760 --> 00:23:16,080
THAT GOES FROM BEING A TETRAMER

681
00:23:16,080 --> 00:23:17,440
ON THE RIGHT IN THE ABSENCE OF

682
00:23:17,440 --> 00:23:21,400
THE PEPTIDE, TO A PENTAMER ON

683
00:23:21,400 --> 00:23:23,240
THE LEFT, WHEN THE PEPTIDE IS

684
00:23:23,240 --> 00:23:23,600
ADDED.

685
00:23:23,600 --> 00:23:24,000
OKAY.

686
00:23:24,000 --> 00:23:25,440
SO NOW I'M GOING TO SWITCH GEARS

687
00:23:25,440 --> 00:23:29,840
AND TALK ABOUT DEEP LEARNING,

688
00:23:29,840 --> 00:23:31,040
AND SO WE'VE DEVELOPED THIS

689
00:23:31,040 --> 00:23:32,760
NETWORK CALLED ROSETTAFOLD,

690
00:23:32,760 --> 00:23:35,640
WHICH IS PRETTY GOOD AT

691
00:23:35,640 --> 00:23:36,680
PREDICTING STRUCTURE FROM

692
00:23:36,680 --> 00:23:37,280
SEQUENCE.

693
00:23:37,280 --> 00:23:39,440
AND IT HAS THREE TRACKS.

694
00:23:39,440 --> 00:23:42,320
IT HAS A TRACK SORT OF SEQUENCE

695
00:23:42,320 --> 00:23:43,280
INFORMATION IS GOING THROUGH ON

696
00:23:43,280 --> 00:23:45,120
THIS TRACK, THEN DISTANCE

697
00:23:45,120 --> 00:23:46,000
INFORMATION IS GOING THROUGH ON

698
00:23:46,000 --> 00:23:47,760
THIS TRACK, AND FINALLY

699
00:23:47,760 --> 00:23:48,520
STRUCTURE INFORMATION IS GOING

700
00:23:48,520 --> 00:23:50,800
THROUGH ON THIS TRACK.

701
00:23:50,800 --> 00:23:52,360
SO VARIOUS TYPES OF INFORMATION

702
00:23:52,360 --> 00:23:53,560
COME IN AND THE STRUCTURE COMES

703
00:23:53,560 --> 00:23:55,280
OUT.

704
00:23:55,280 --> 00:23:58,680
SO ONE OF THE THINGS THAT WE'VE

705
00:23:58,680 --> 00:24:01,840
BEEN DOING RECENTLY IS TO EXTEND

706
00:24:01,840 --> 00:24:03,920
TO MORE GENERAL PROBLEMS, SO FOR

707
00:24:03,920 --> 00:24:04,840
EXAMPLE, WE NOW HAVE A VERSION

708
00:24:04,840 --> 00:24:08,880
OF THE NETWORK THAT IS WORK FROM

709
00:24:08,880 --> 00:24:12,800
FRANK DEMAIO THAT WILL TAKE IN

710
00:24:12,800 --> 00:24:15,640
NOT ONLY PROTEIN SEQUENCE BUT

711
00:24:15,640 --> 00:24:17,320
ALSO DNA AND RNA SEQUENCE, SO

712
00:24:17,320 --> 00:24:21,160
MODELS OF PROTEIN RNA OR PROTEIN

713
00:24:21,160 --> 00:24:24,960
DNDNA MODELS AND THESE ARE

714
00:24:24,960 --> 00:24:26,600
EXAMPLES OF SOME OF THE

715
00:24:26,600 --> 00:24:31,720
PREDICTIONED PPREDICTED PROTEINS

716
00:24:31,720 --> 00:24:33,160
THAT IT HAS GENERATED.

717
00:24:33,160 --> 00:24:34,680
I APOLOGIZE FOR HOW COMPLICATED

718
00:24:34,680 --> 00:24:35,320
THIS SLIDE IS.

719
00:24:35,320 --> 00:24:37,520
THIS IS SHOWING YOU WHAT THE

720
00:24:37,520 --> 00:24:38,480
ARCHITECTURE OF ONE OF THESE

721
00:24:38,480 --> 00:24:39,840
NEURAL NETWORKS LOOKS LIKE.

722
00:24:39,840 --> 00:24:40,960
IF YOU HAVEN'T SEEN A LOT OF

723
00:24:40,960 --> 00:24:42,440
THESE BEFORE, THEY'RE KIND OF

724
00:24:42,440 --> 00:24:43,880
COMPLICATED AND DON'T WORRY TOO

725
00:24:43,880 --> 00:24:46,280
MUCH ABOUT THE DETAILS, BUT THIS

726
00:24:46,280 --> 00:24:47,920
DOES THE CLASSIC DESIGN PROGRAM

727
00:24:47,920 --> 00:24:50,120
OF GIVEN THE BACKBONE OF A

728
00:24:50,120 --> 00:24:52,760
PROTEIN, PREDICT WHAT AMINO ACID

729
00:24:52,760 --> 00:24:53,920
SEQUENCES WHICH WILL FOLD TO

730
00:24:53,920 --> 00:24:55,360
THAT BACKBONE.

731
00:24:55,360 --> 00:24:56,640
AND ONE OF THE THINGS WE FOUND,

732
00:24:56,640 --> 00:24:58,080
WE'VE BEEN DEVELOPING THIS

733
00:24:58,080 --> 00:25:00,240
ROSETTA PHYSICALLY-BASED METHOD

734
00:25:00,240 --> 00:25:01,680
FOR MANY YEARS, AND WE FOUND

735
00:25:01,680 --> 00:25:03,360
THAT WHEN WE DEVELOP THIS NEW

736
00:25:03,360 --> 00:25:05,840
METHOD CALLED PROTEIN PMNN AND

737
00:25:05,840 --> 00:25:07,600
WE DID THE CLASSIC TEST OF

738
00:25:07,600 --> 00:25:09,120
TAKING MANY, MANY NATURALLY

739
00:25:09,120 --> 00:25:12,000
OCCURRING PROTEIN STRUCTURES,

740
00:25:12,000 --> 00:25:12,760
THROWING AWAY THE SIDE CHAINS

741
00:25:12,760 --> 00:25:15,080
AND ASKING THE NETWORK TO PRO

742
00:25:15,080 --> 00:25:16,520
DICTIONARY WHAT THE SEQUENCE OF

743
00:25:16,520 --> 00:25:19,880
THE PROTEIN IS THAT THIS

744
00:25:19,880 --> 00:25:22,720
NETWORK DID FAR BETTER, SO THE

745
00:25:22,720 --> 00:25:24,000
IT PREDICTS THE SEQUENCE BASED

746
00:25:24,000 --> 00:25:28,720
ON THE STRUCTURE, AND THEY DID

747
00:25:28,720 --> 00:25:30,360
A REALLY GOOD JOB AT PREDICTING

748
00:25:30,360 --> 00:25:32,800
WHAT THE NATIVE AMINO ACID

749
00:25:32,800 --> 00:25:37,320
SEQUENCES WERE FAR BETTER THAN

750
00:25:37,320 --> 00:25:37,560
ROSETTA.

751
00:25:37,560 --> 00:25:40,840
SO THAT BECAME VERY EVIDENT IN

752
00:25:40,840 --> 00:25:42,480
THE WORK LEADING UP TO THIS

753
00:25:42,480 --> 00:25:42,960
EXPERIMENT HERE.

754
00:25:42,960 --> 00:25:45,160
SO HERE WHAT WE DID IS, IT WAS

755
00:25:45,160 --> 00:25:47,880
OUR FIRST ATTEMPT TO DESIGN

756
00:25:47,880 --> 00:25:49,640
PROTEINS USING ONE OF THESE

757
00:25:49,640 --> 00:25:50,120
NETWORKS.

758
00:25:50,120 --> 00:25:51,600
SO WE STARTED WITH COMPLETELY

759
00:25:51,600 --> 00:25:53,920
RANDOM SEQUENCES AND WE

760
00:25:53,920 --> 00:25:54,560
PREDICTED THE STRUCTURE.

761
00:25:54,560 --> 00:25:56,640
IN THIS CASE, WE USED ALPHA

762
00:25:56,640 --> 00:25:57,960
FOLD, AND A COMPLETELY RANDOM

763
00:25:57,960 --> 00:25:59,360
SEQUENCE WON'T BE PREDICTED TO

764
00:25:59,360 --> 00:26:00,120
HAVE ANY PARTICULAR STRUCTURE AT

765
00:26:00,120 --> 00:26:00,440
ALL.

766
00:26:00,440 --> 00:26:02,320
BUT THEN WE COULD START CHANGING

767
00:26:02,320 --> 00:26:03,760
THE SEQUENCE AND IDENTIFY

768
00:26:03,760 --> 00:26:05,360
CHANGES, BASICALLY DO A MONTE

769
00:26:05,360 --> 00:26:06,680
CARLO SEARCH, IDENTIFY SEQUENCE

770
00:26:06,680 --> 00:26:08,200
CHANGES THAT MADE IT PREDICTED

771
00:26:08,200 --> 00:26:14,360
TO FORM SOME STRUCTURE, AND THEN

772
00:26:14,360 --> 00:26:15,760
WE WOULD GET SEQUENCES STRONGLY

773
00:26:15,760 --> 00:26:16,760
PREDICTED TO FOLD TO A

774
00:26:16,760 --> 00:26:17,240
STRUCTURE.

775
00:26:17,240 --> 00:26:20,480
SO IN THIS EXPERIMENT, WHAT WE

776
00:26:20,480 --> 00:26:23,320
DID WAS TO ASK NOT ONLY THAT THE

777
00:26:23,320 --> 00:26:25,600
SEQUENCE FOLD TO A STRUCTURE BUT

778
00:26:25,600 --> 00:26:27,560
THAT IT WOULD BE TO A STRUCTURE.

779
00:26:27,560 --> 00:26:28,920
NOW WHEN WE DID THIS WE GOT THE

780
00:26:28,920 --> 00:26:30,120
SEQUENCES OUT THAT WERE

781
00:26:30,120 --> 00:26:32,960
PREDICTED BY ALPHAFOLD TO FOLD

782
00:26:32,960 --> 00:26:34,800
TO A VERY NICE OLIGAMERIC

783
00:26:34,800 --> 00:26:35,920
STRUCTURE, BUT WHEN WE MADE THEM

784
00:26:35,920 --> 00:26:38,960
IN THE LAB, AND THE WAY WE'VE

785
00:26:38,960 --> 00:26:41,000
DONE -- EVERYTHING HAS BEEN DONE

786
00:26:41,000 --> 00:26:43,200
THAT I'VE DESCRIBED SO FAR IS WE

787
00:26:43,200 --> 00:26:45,080
GET SYNTHETIC GENES, PUT THEM

788
00:26:45,080 --> 00:26:46,720
INTO BACTERIA, BACTERIA MAKE THE

789
00:26:46,720 --> 00:26:48,040
PROTEIN AND IN THE EXAMPLES I

790
00:26:48,040 --> 00:26:49,840
SHOWED YOU SO FAR, WE GOT THESE

791
00:26:49,840 --> 00:26:51,720
NICE PROTEINS THAT WERE FLU

792
00:26:51,720 --> 00:26:52,640
INHIBITORS OR WHATEVER.

793
00:26:52,640 --> 00:26:55,880
SO IN THIS CASE WE JUST GOT GUNK

794
00:26:55,880 --> 00:26:57,080
AT THE BOTTOM OF THE TUBE.

795
00:26:57,080 --> 00:27:00,400
THEN WE TOOK THE PROTEIN MPMN,

796
00:27:00,400 --> 00:27:01,800
FOUND A NEW SEQUENCE, IN THAT

797
00:27:01,800 --> 00:27:03,440
CASE, THE SUCCESS RATE WAS VERY

798
00:27:03,440 --> 00:27:08,720
HIGH AND WE WERE ABLE TO GET

799
00:27:08,720 --> 00:27:10,560
ULTIMATELY THESE CRYO-EM

800
00:27:10,560 --> 00:27:11,640
STRUCTURES OF THESE VERY LARGE

801
00:27:11,640 --> 00:27:12,760
RINGS THAT AGAIN JUST CAME OUT

802
00:27:12,760 --> 00:27:15,920
FROM THIS RANDOM SEQUENCE

803
00:27:15,920 --> 00:27:16,360
SEARCH.

804
00:27:16,360 --> 00:27:19,240
SO THAT WAS SORT OF OUR FIRST

805
00:27:19,240 --> 00:27:20,200
EXPERIMENTAL EVIDENCE THAT THIS

806
00:27:20,200 --> 00:27:22,920
PROTEIN MPNN COULD REALLY MAKE

807
00:27:22,920 --> 00:27:23,440
PROTEINS THAT FOLD.

808
00:27:23,440 --> 00:27:25,000
AND THESE ARE VERY CLOSE TO THE

809
00:27:25,000 --> 00:27:26,520
DESIGN MODELS.

810
00:27:26,520 --> 00:27:29,160
SO THEN ANOTHER EXAMPLE OF USING

811
00:27:29,160 --> 00:27:31,920
DEEP LEARNING IS ENZYMES.

812
00:27:31,920 --> 00:27:33,320
SO DESIGNING ENZYMES IS HARDER

813
00:27:33,320 --> 00:27:34,640
THAN WHAT I'VE DESCRIBED SO FAR

814
00:27:34,640 --> 00:27:35,720
BECAUSE IT'S NOT JUST BINDING.

815
00:27:35,720 --> 00:27:38,120
YOU HAVE TO SELECTIVELY

816
00:27:38,120 --> 00:27:39,360
STABILIZE THE TRANSITION STATE.

817
00:27:39,360 --> 00:27:41,600
AND THE REACTION THAT ANDY AND

818
00:27:41,600 --> 00:27:45,200
CHRIS CHOSE WAS THIS LUCIFERASE

819
00:27:45,200 --> 00:27:48,520
REACTION, AND THEY MADE PROTEINS

820
00:27:48,520 --> 00:27:50,040
WHICH SELECTIVELY STABILIZE THE

821
00:27:50,040 --> 00:27:51,000
TRANSITION STATE FOR THIS

822
00:27:51,000 --> 00:27:51,840
REACTION.

823
00:27:51,840 --> 00:27:53,120
AND THE WAY THEY DID THIS WAS TO

824
00:27:53,120 --> 00:27:54,600
USE THESE DEEP LEARNING METHODS,

825
00:27:54,600 --> 00:27:57,040
WE CALL IT HALLUCINATION, THE

826
00:27:57,040 --> 00:27:59,120
METHOD I DESCRIBED JUST A MOMENT

827
00:27:59,120 --> 00:27:59,880
AGO.

828
00:27:59,880 --> 00:28:01,760
WE'RE SEARCHING OVER ALL THE

829
00:28:01,760 --> 00:28:02,880
POSSIBLE INPUTS OF THE NETWORK,

830
00:28:02,880 --> 00:28:04,080
IT JUST KIND OF MAKING UP THINGS

831
00:28:04,080 --> 00:28:04,680
FROM SCRATCH.

832
00:28:04,680 --> 00:28:06,760
AND WE SEARCHED FOR SEQUENCES

833
00:28:06,760 --> 00:28:08,320
THAT FOLD UP TO SCAFFOLDS THAT

834
00:28:08,320 --> 00:28:09,720
LOOK LIKE THIS.

835
00:28:09,720 --> 00:28:12,360
AND THEN WE -- IN THIS CASE THIS

836
00:28:12,360 --> 00:28:14,120
WAS SORT OF A HYBRID, WE USED

837
00:28:14,120 --> 00:28:15,560
ROSETTA TO DESIGN SIDE CHAINS

838
00:28:15,560 --> 00:28:16,600
THAT INTERACT WITH THIS SUBTRAIT

839
00:28:16,600 --> 00:28:18,480
FOR THE LUCIFERASE REACTION.

840
00:28:18,480 --> 00:28:20,000
AND WHAT WE FOUND IS WE GOT

841
00:28:20,000 --> 00:28:22,640
QUITE ACTIVE LUCIFERASES OUT OF

842
00:28:22,640 --> 00:28:24,480
THIS, AND THEY WERE CLOSE TO AS

843
00:28:24,480 --> 00:28:27,440
ACTIVE AS NATURALLY OCCURRING

844
00:28:27,440 --> 00:28:28,800
LUCIFERASES FOR THE

845
00:28:28,800 --> 00:28:30,280
ENZYMOLOGISTS WHO MIGHT BE

846
00:28:30,280 --> 00:28:32,480
LISTENING, 10 TO THE SIXTH PER

847
00:28:32,480 --> 00:28:34,040
MOLD PER SECOND, WHICH IS REALLY

848
00:28:34,040 --> 00:28:34,560
PRETTY ACTIVE.

849
00:28:34,560 --> 00:28:36,760
THERE WAS AN INTERESTING

850
00:28:36,760 --> 00:28:37,960
DIFFERENCE FROM -- SORRY.

851
00:28:37,960 --> 00:28:40,080
I KEEP DOING THAT.

852
00:28:40,080 --> 00:28:45,080
FROM NATURALLY OCCURRING

853
00:28:45,080 --> 00:28:46,800
LUCIFERASES, WHICH HAVE VERY

854
00:28:46,800 --> 00:28:49,160
BROAD SUBSTRATE SPECIFICITY.

855
00:28:49,160 --> 00:28:50,760
THIS DESIGN LUCIFERASE IS VERY

856
00:28:50,760 --> 00:28:51,400
HIGHLY SPECIFIC BECAUSE THE

857
00:28:51,400 --> 00:28:52,640
POCKET IS VERY WELL PACKED

858
00:28:52,640 --> 00:28:54,600
AROUND THE SUBSTRATE.

859
00:28:54,600 --> 00:28:56,440
SO ANDY IS NOW TRYING TO MAKE

860
00:28:56,440 --> 00:28:57,320
LUCIFERASES THAT ARE SPECIFIC

861
00:28:57,320 --> 00:28:58,760
FOR OTHER SUBSTRATE.

862
00:28:58,760 --> 00:28:59,280
OKAY.

863
00:28:59,280 --> 00:29:01,240
SO NOW THE REST OF MY TALK, I'M

864
00:29:01,240 --> 00:29:02,800
GOING TO TALK ABOUT REALLY WORK

865
00:29:02,800 --> 00:29:04,440
OF THE LAST THREE MONTH THAT'S

866
00:29:04,440 --> 00:29:08,360
INVOLVED A LOT OF PEOPLE AND IT

867
00:29:08,360 --> 00:29:09,880
VERY EXCITING, SO I HOPE YOU

868
00:29:09,880 --> 00:29:13,320
WILL ENJOY THIS.

869
00:29:13,320 --> 00:29:14,720
SO THIS HALLUCINATION APPROACH

870
00:29:14,720 --> 00:29:16,360
WHERE WE WERE KIND OF JUST

871
00:29:16,360 --> 00:29:18,360
RANDOMLY SEARCHING IN SEQUENCE

872
00:29:18,360 --> 00:29:19,760
SPACE FOR SEQUENCES THAT WILL BE

873
00:29:19,760 --> 00:29:22,160
PREDICTED BY THE NETWORK TO HAVE

874
00:29:22,160 --> 00:29:23,360
THE PROPERTIES THAT WE WANTED

875
00:29:23,360 --> 00:29:25,000
WAS NOT VERY EFFICIENT BECAUSE

876
00:29:25,000 --> 00:29:26,480
EVERY TIME WE CHANGED THE

877
00:29:26,480 --> 00:29:28,200
SEQUENCE, WE HAD TO MAKE A

878
00:29:28,200 --> 00:29:30,240
STRUCTURE PREDICTION AGAIN WITH

879
00:29:30,240 --> 00:29:31,200
ROSETTAFOLD OR ALPHAFOLD.

880
00:29:31,200 --> 00:29:32,560
SO WE COULDN'T REALLY DIRECT IT.

881
00:29:32,560 --> 00:29:33,840
SO WE WANTED REALLY A MORE

882
00:29:33,840 --> 00:29:35,160
DIRECTED WAY OF DOING THIS.

883
00:29:35,160 --> 00:29:38,040
SO THE FIRST WAY WE TRIED TO DO

884
00:29:38,040 --> 00:29:40,200
THIS IS, WE SORT OF TOOK

885
00:29:40,200 --> 00:29:41,520
ROSETTAFOLD AND TREATED IT

886
00:29:41,520 --> 00:29:42,600
LIKE -- WELL, YOU KNOW HOW WHEN

887
00:29:42,600 --> 00:29:44,680
YOU TYPE A FEW WORDS INTO -- ON

888
00:29:44,680 --> 00:29:46,120
A WORD PROCESSOR, IT WILL TRY

889
00:29:46,120 --> 00:29:47,120
AND SOMETIMES COMPLETE YOUR

890
00:29:47,120 --> 00:29:48,520
SENTENCE, SORT OF THAT AUTO

891
00:29:48,520 --> 00:29:49,920
COMPLEEX.

892
00:29:49,920 --> 00:29:50,680
COMPLETION.

893
00:29:50,680 --> 00:29:53,120
WE TRIED TO USE ROSETTAFOLD AS

894
00:29:53,120 --> 00:29:54,400
SORT OF A PROTEIN AUTO

895
00:29:54,400 --> 00:29:54,960
COMPLETER.

896
00:29:54,960 --> 00:29:55,960
WHAT WE DID IS RATHER THAN JUST

897
00:29:55,960 --> 00:29:56,960
TRAINING IT ON SEQUENCE

898
00:29:56,960 --> 00:29:59,240
INFORMATION AND ASKING TO

899
00:29:59,240 --> 00:30:00,760
PREDICT STRUCTURES, SOME OF THE

900
00:30:00,760 --> 00:30:02,000
TIME WE WOULD DELETE SEQUENCE

901
00:30:02,000 --> 00:30:02,720
AND STRUCTURE INFORMATION AND

902
00:30:02,720 --> 00:30:06,120
ASK TO COMPLETE THE MISSING

903
00:30:06,120 --> 00:30:06,640
SEQUENCE AND STRUCTURAL

904
00:30:06,640 --> 00:30:08,520
INFORMATION.

905
00:30:08,520 --> 00:30:10,160
SO THEN WHAT WE COULD DO IS TAKE

906
00:30:10,160 --> 00:30:12,360
JUST PARTS OF PROTEINS, LIKE IN

907
00:30:12,360 --> 00:30:15,680
THIS CASE A DIIRON SITE, WE

908
00:30:15,680 --> 00:30:17,160
COULD JUST TAKE A LITTLE BIT OF

909
00:30:17,160 --> 00:30:18,760
A PROTEIN YOU OUT OF FUNCTION,

910
00:30:18,760 --> 00:30:20,320
LEAVE OUT THE REST OF THE

911
00:30:20,320 --> 00:30:22,520
SEQUENCE AND STRUCTURE AND

912
00:30:22,520 --> 00:30:23,400
ROSETTAFOLD WOULD JUST BUILD IT

913
00:30:23,400 --> 00:30:23,840
IN.

914
00:30:23,840 --> 00:30:27,000
IN THAT WAY, WE WERE ABLE TO GET

915
00:30:27,000 --> 00:30:28,760
PROTEINS THAT IS A CRYSTAL

916
00:30:28,760 --> 00:30:33,800
STRUCTURE ACTUALLY HAD THAT

917
00:30:33,800 --> 00:30:35,120
DESIGN SITE AND BOUND METAL.

918
00:30:35,120 --> 00:30:37,440
WE COULD ALSO -- WE'VE BEEN

919
00:30:37,440 --> 00:30:38,840
OBVIOUSLY INTERESTED IN BETTER

920
00:30:38,840 --> 00:30:40,400
WAYS OF MAKING VACCINES, WE

921
00:30:40,400 --> 00:30:42,120
COULD TAKE EPITOPES FROM A VIRUS

922
00:30:42,120 --> 00:30:45,280
LIKE IN THIS CASE RESPIRATORY

923
00:30:45,280 --> 00:30:49,120
SYNCYTIAL VIRUS, RSVF PROTEIN,

924
00:30:49,120 --> 00:30:50,360
WE COULD TAKE EPITOPES AND ASK

925
00:30:50,360 --> 00:30:51,960
FOR A SINGLE PROTEIN THAT COULD

926
00:30:51,960 --> 00:30:53,080
PRESENT THEM ALL IN THEIR FOLDED

927
00:30:53,080 --> 00:30:53,400
FORM.

928
00:30:53,400 --> 00:30:58,560
SO HERE ARE A COUPLE OF THESE

929
00:30:58,560 --> 00:31:00,760
INPATED DESIGNS MADE USING THIS

930
00:31:00,760 --> 00:31:01,000
APPROACH.

931
00:31:01,000 --> 00:31:02,680
THEY ACTUALLY BIND ALL THREE

932
00:31:02,680 --> 00:31:04,360
DIFFERENT NEUTRALIZING

933
00:31:04,360 --> 00:31:05,120
ANTIBODIES AGAINST THESE THREE

934
00:31:05,120 --> 00:31:05,640
DIFFERENT SITES.

935
00:31:05,640 --> 00:31:08,520
SO WE'RE VERY EXCITED NOW TO SEE

936
00:31:08,520 --> 00:31:11,680
HOW THEY WORK AS IMMUNOGENS.

937
00:31:11,680 --> 00:31:12,640
BUT THERE WAS A PROBLEM, AND

938
00:31:12,640 --> 00:31:16,160
THAT WAS THAT THIS INPAINTING

939
00:31:16,160 --> 00:31:17,080
APPROACH DOES EVERYTHING IN ONE

940
00:31:17,080 --> 00:31:17,400
STEP.

941
00:31:17,400 --> 00:31:20,000
SO WE COULDN'T GET IT, SO IT

942
00:31:20,000 --> 00:31:21,680
ALWAYS -- IT WAS LIKE HAVING A

943
00:31:21,680 --> 00:31:23,080
SENTENCE AUTO COMPLETED THAT

944
00:31:23,080 --> 00:31:24,160
ALWAYS AUTO COMPLETED THE SAME

945
00:31:24,160 --> 00:31:26,120
WAY, AND IN THE SAME WAY THAT A

946
00:31:26,120 --> 00:31:28,440
SENTENCE AUTO COMPLETER WOULD

947
00:31:28,440 --> 00:31:29,640
HAVE, YOU ALWAYS HAD TO GIVE IT

948
00:31:29,640 --> 00:31:30,720
SOME MINIMAL AMOUNT OF

949
00:31:30,720 --> 00:31:31,240
INFORMATION.

950
00:31:31,240 --> 00:31:33,000
YOU GAVE IT TOO SMALL A

951
00:31:33,000 --> 00:31:34,640
FUNCTIONAL MOTIF, IT JUST

952
00:31:34,640 --> 00:31:36,840
COULDN'T FILL IN THE REST.

953
00:31:36,840 --> 00:31:38,600
SO DIFFUSION METHODS HAVE BECOME

954
00:31:38,600 --> 00:31:42,200
VERY POPULAR FOR THINGS LIKE --

955
00:31:42,200 --> 00:31:44,200
FOR BUILDING IMAGES AND THINGS

956
00:31:44,200 --> 00:31:46,400
LIKE DOLLY THAT GO FROM TEXT TO

957
00:31:46,400 --> 00:31:48,040
IMAGES ALL USE DIFFUSION MODELS.

958
00:31:48,040 --> 00:31:50,760
AND THE BASIC WAY THEY WORK IS

959
00:31:50,760 --> 00:31:53,840
YOU START WITH A COMPLETELY

960
00:31:53,840 --> 00:31:54,600
RANDOM NOISE DISTRIBUTION AS

961
00:31:54,600 --> 00:31:57,080
SHOWN ON THE LEFT.

962
00:31:57,080 --> 00:31:59,680
AND THEN YOU SUCCESSIVELY REMOVE

963
00:31:59,680 --> 00:32:02,120
THE NOISE AND YOU DO THAT AT

964
00:32:02,120 --> 00:32:03,480
EACH STEP, YOU REMOVE A SMALL

965
00:32:03,480 --> 00:32:06,960
AMOUNT OF THE NOISE.

966
00:32:06,960 --> 00:32:08,040
AND WHAT WE'RE USING FOR

967
00:32:08,040 --> 00:32:10,440
REMOVING THAT NOISE IS ROSETTA

968
00:32:10,440 --> 00:32:10,920
FOLD.

969
00:32:10,920 --> 00:32:14,080
WE ASK -- WE GIVE ROSETTA FOLD,

970
00:32:14,080 --> 00:32:15,600
WE START WITH COMPLETELY RANDOM

971
00:32:15,600 --> 00:32:17,160
NOISE, WE ASK TO PREDICT WHAT

972
00:32:17,160 --> 00:32:20,800
THE STRUCTURE IS, IT WILL MAKE

973
00:32:20,800 --> 00:32:22,000
SOME GUESS AND THEN WE TAKE A

974
00:32:22,000 --> 00:32:23,400
SMALL STEP IN THAT DIRECTION,

975
00:32:23,400 --> 00:32:25,040
AND THAT'S HOW WE MOVE DOWN TO

976
00:32:25,040 --> 00:32:27,800
LESS AND LESS NOISE.

977
00:32:27,800 --> 00:32:31,840
THEN AS WE GET CLOSE E  CLOSER E

978
00:32:31,840 --> 00:32:32,920
REMOVE MORE AND MORE NOISE IT

979
00:32:32,920 --> 00:32:34,280
STARTS TO LOOK MORE AND MORE

980
00:32:34,280 --> 00:32:37,560
LIKE A PROTEIN STRUCTURE.

981
00:32:37,560 --> 00:32:38,760
IN THE LAST FEW MINUTES, I'LL

982
00:32:38,760 --> 00:32:40,160
SHOW YOU EXAMPLES OF HOW WE CAN

983
00:32:40,160 --> 00:32:41,720
APPLY THIS FOR A NUMBER OF

984
00:32:41,720 --> 00:32:46,200
DIFFERENT TYPES OF DESIGN

985
00:32:46,200 --> 00:32:46,840
PROBLEMS.

986
00:32:46,840 --> 00:32:48,040
THE FIRST THING WE CAN DO IS USE

987
00:32:48,040 --> 00:32:49,920
THIS TO MAKE UP BRAND NEW

988
00:32:49,920 --> 00:32:50,480
PROTEIN STRUCTURES.

989
00:32:50,480 --> 00:32:52,240
SO WE START WITH RANDOM NOISE

990
00:32:52,240 --> 00:32:55,200
AND PUT IT THROUGH THIS

991
00:32:55,200 --> 00:32:56,040
DENOISING PROCESS AND EVERY TIME

992
00:32:56,040 --> 00:32:57,520
WE DO THIS, WE GET A DIFFERENT

993
00:32:57,520 --> 00:32:58,880
STRUCTURE OUT.

994
00:32:58,880 --> 00:33:00,960
THESE ARE SOME EXAMPLES OF

995
00:33:00,960 --> 00:33:02,200
ACTUALLY VERY LARGE PROTEINS WE

996
00:33:02,200 --> 00:33:04,280
CAN GET THIS WAY THAT LOOK LIKE

997
00:33:04,280 --> 00:33:06,040
THEY COULD BE PROTEINS, BUT THEY

998
00:33:06,040 --> 00:33:08,040
AREN'T ANY ACTUALLY OCCURRING

999
00:33:08,040 --> 00:33:11,600
PROTEIN.

1000
00:33:11,600 --> 00:33:14,200
NOW, WHAT WE CAN ALSO DO IS WE

1001
00:33:14,200 --> 00:33:15,560
CAN TAKE THIS RANDOM NOISE AND

1002
00:33:15,560 --> 00:33:17,320
THEN WE CAN MAKE COPIES OF IT,

1003
00:33:17,320 --> 00:33:18,960
IN THIS CASE THREE COPIES, AND

1004
00:33:18,960 --> 00:33:20,240
GO THROUGH THE DENOISING

1005
00:33:20,240 --> 00:33:21,880
PROCESS, FORCING THE NOISE TO BE

1006
00:33:21,880 --> 00:33:23,960
SYMMETRIC AT EACH STAGE.

1007
00:33:23,960 --> 00:33:27,440
SO WHAT HAPPENS THEN IS WE GET

1008
00:33:27,440 --> 00:33:28,640
SYMMETRIC OLIGAMERIC STRUCTURES

1009
00:33:28,640 --> 00:33:32,640
OUT, AS YOU CAN SEE HERE.

1010
00:33:32,640 --> 00:33:33,880
I'LL JUST LET THAT PLAY ONE MORE

1011
00:33:33,880 --> 00:33:35,880
TIME.

1012
00:33:35,880 --> 00:33:37,440
SO BECAUSE WE'RE STARTING WITH

1013
00:33:37,440 --> 00:33:38,320
THE RANDOM NOISE DISTRIBUTION

1014
00:33:38,320 --> 00:33:39,840
EVERY TIME WE RUN THIS, WE GET

1015
00:33:39,840 --> 00:33:41,360
SOMETHING DIFFERENT OUT.

1016
00:33:41,360 --> 00:33:43,000
AND SO HERE ARE SOME OF THE

1017
00:33:43,000 --> 00:33:44,120
THINGS THAT COME OUT.

1018
00:33:44,120 --> 00:33:47,520
AND IT REALLY QUITE FASCINATING.

1019
00:33:47,520 --> 00:33:48,480
SOMETIMES -- THESE ARE JUST

1020
00:33:48,480 --> 00:33:50,640
DIFFERENT EXAMPLES AND WHAT YOU

1021
00:33:50,640 --> 00:33:55,040
SEE HERE IS EM DATA ON THESE

1022
00:33:55,040 --> 00:33:55,360
STRUCTURES.

1023
00:33:55,360 --> 00:34:00,520
THIS IS WORK FROM HELEN ISANAC

1024
00:34:00,520 --> 00:34:02,040
AND ANDREW -- YOU CAN SEE THEY

1025
00:34:02,040 --> 00:34:02,480
CLOSE.

1026
00:34:02,480 --> 00:34:03,680
FOR THOSE FAMILIAR WITH NATURAL

1027
00:34:03,680 --> 00:34:07,760
PROTEINS KNOW THAT -- WHAT THIS

1028
00:34:07,760 --> 00:34:09,080
NETWORK IS DOING IS GENERATING

1029
00:34:09,080 --> 00:34:11,240
THINGS THAT LOOK LIKE TIN

1030
00:34:11,240 --> 00:34:13,760
BARRELS BUT THEY HAVE 24 STRANDS

1031
00:34:13,760 --> 00:34:18,040
AND 24 HELICES -- SO THE NETWORK

1032
00:34:18,040 --> 00:34:19,360
HAS BEEN TRAINED ON PROTEIN

1033
00:34:19,360 --> 00:34:21,120
STRUCTURES BUT NOW IT CAN GO

1034
00:34:21,120 --> 00:34:21,560
CONSIDERABLY BEYOND.

1035
00:34:21,560 --> 00:34:24,080
SO WE CAN ALSO ASK TO MAKE, JUST

1036
00:34:24,080 --> 00:34:28,200
BY THE WAY IN WHICH WE'RE

1037
00:34:28,200 --> 00:34:29,880
SYMMETRI SOO ING THE NOISE, WE

1038
00:34:29,880 --> 00:34:32,200
CAN ASK TO MAKE DIE HEE DRAL

1039
00:34:32,200 --> 00:34:32,480
STRUCTURES.

1040
00:34:32,480 --> 00:34:36,320
SO IT'S GOT LIKE TWO CYCLIC

1041
00:34:36,320 --> 00:34:38,240
RINGS OF FOUR SUBUNITS EACH

1042
00:34:38,240 --> 00:34:39,000
STACKED ON EACH OTHER.

1043
00:34:39,000 --> 00:34:40,520
HERE YOU CAN SEE THE EM

1044
00:34:40,520 --> 00:34:42,080
RECONSTRUCTION IS VERY CLOSE TO

1045
00:34:42,080 --> 00:34:47,760
THE DESIGN MODEL.

1046
00:34:47,760 --> 00:34:49,200
COMING BACK TO THIS VERY FIRST

1047
00:34:49,200 --> 00:34:50,880
SLIDE I SHOWED WITH THE COVID

1048
00:34:50,880 --> 00:34:53,400
VACCINE, WE CAN JUST GIVE IT A

1049
00:34:53,400 --> 00:34:53,920
CERTAIN CHAIN LENGTH.

1050
00:34:53,920 --> 00:34:56,960
IN ALL OF THESE CASES WE HAVE TO

1051
00:34:56,960 --> 00:35:01,000
SPECIFY THE CHAIN LENGTH AND SAY

1052
00:35:01,000 --> 00:35:04,000
MAKE SOMETHING THAT'S

1053
00:35:04,000 --> 00:35:04,320
ICOSAHEDRAL.

1054
00:35:04,320 --> 00:35:06,200
THIS IS WHAT THE PROTEINS LOOK

1055
00:35:06,200 --> 00:35:09,360
LIKE, THESE NICE LITTLE

1056
00:35:09,360 --> 00:35:12,080
ICOSAHEDRA, AND THEN THESE ARE

1057
00:35:12,080 --> 00:35:13,640
THE NEGATIVE STAIN CLASS

1058
00:35:13,640 --> 00:35:18,240
AVERAGES AGAIN MATCH THE -- OR

1059
00:35:18,240 --> 00:35:19,200
RECONSTRUCTIONS MATCH THE MODEL.

1060
00:35:19,200 --> 00:35:20,440
SO WHAT ABOUT BINDING?

1061
00:35:20,440 --> 00:35:21,520
I PROMISED TO GET BACK TO

1062
00:35:21,520 --> 00:35:21,840
BINDING.

1063
00:35:21,840 --> 00:35:24,360
NOW THIS IS THE VERY LAST FEW

1064
00:35:24,360 --> 00:35:24,720
SLIDES.

1065
00:35:24,720 --> 00:35:26,440
SO THE FIRST PROBLEM WE CAN DO

1066
00:35:26,440 --> 00:35:30,840
IS, THIS IS MDM2, WHICH IS --

1067
00:35:30,840 --> 00:35:33,360
WHICH DEGRADES P53, SO IT'S

1068
00:35:33,360 --> 00:35:35,320
IMPORTANT TUMOR SUPPRESSOR.

1069
00:35:35,320 --> 00:35:37,960
WE CAN TAKE JUST THE P53 PEPTIDE

1070
00:35:37,960 --> 00:35:40,240
AND SAY, BUILD A PROTEIN AROUND

1071
00:35:40,240 --> 00:35:42,560
IT THAT BINDS MDM2.

1072
00:35:42,560 --> 00:35:45,200
AND WHEN WE DO THAT, WE GET OUT

1073
00:35:45,200 --> 00:35:47,120
PROTEINS THAT BIND WITH VERY

1074
00:35:47,120 --> 00:35:48,720
HIGH AFFINITY TO MDM2.

1075
00:35:48,720 --> 00:35:50,120
WE DIDN'T HAVE TO -- THE

1076
00:35:50,120 --> 00:35:51,560
PROTEINS BEING MADE IN PLACE IN

1077
00:35:51,560 --> 00:35:53,760
THIS CASE, IT'S NOT -- IN THE

1078
00:35:53,760 --> 00:35:54,560
CALCULATIONS I TALKED ABOUT IN

1079
00:35:54,560 --> 00:35:55,840
THE BEGINNING OF MY TALK, WE

1080
00:35:55,840 --> 00:35:57,160
STARTED BY GENERATING A VERY

1081
00:35:57,160 --> 00:35:59,240
LARGE SET OF DIFFERENT SHAPED

1082
00:35:59,240 --> 00:36:00,360
SCAFFOLDS, THEN WE DOCKED THEM

1083
00:36:00,360 --> 00:36:01,560
AGAINST THE TARGET.

1084
00:36:01,560 --> 00:36:03,160
HERE WE'RE JUST STARTING WITH

1085
00:36:03,160 --> 00:36:04,720
RANDOM NOISE PLACED AGAINST THE

1086
00:36:04,720 --> 00:36:06,040
TARGET AND THEN IT SORT OF

1087
00:36:06,040 --> 00:36:08,120
ASSEMBLES INTO THIS PROTEIN THAT

1088
00:36:08,120 --> 00:36:10,840
INCLUDES THE P53 SEGMENT.

1089
00:36:10,840 --> 00:36:12,200
AND NOW FOR THAT PROBLEM I

1090
00:36:12,200 --> 00:36:14,160
DESCRIBED AT THE BEGINNING OF

1091
00:36:14,160 --> 00:36:16,440
HOW DO YOU CONNECT IF YOU HAVE A

1092
00:36:16,440 --> 00:36:18,840
DESIGN PROTEIN BOUND TO A VIRUS,

1093
00:36:18,840 --> 00:36:20,400
CAN YOU CONNECT THOSE PROTEINS

1094
00:36:20,400 --> 00:36:22,080
IN A MEANINGFUL WAY?

1095
00:36:22,080 --> 00:36:24,880
NOW WE CAN DO THIS WITH THIS

1096
00:36:24,880 --> 00:36:26,560
VERY QUICK DEEP LEARNING METHOD,

1097
00:36:26,560 --> 00:36:28,000
WHERE IT'S ESSENTIALLY BUILDING

1098
00:36:28,000 --> 00:36:31,160
A SYMMETRIC TRIMER, BUILDING A

1099
00:36:31,160 --> 00:36:32,240
SYMMETRIC SCAFFOLD IN PLACE

1100
00:36:32,240 --> 00:36:34,760
THAT'S PERFECT FOR HOLDING THESE

1101
00:36:34,760 --> 00:36:40,240
DESIGN BINDERS ON.

1102
00:36:40,240 --> 00:36:41,360
SO SOME EXPERIMENTAL

1103
00:36:41,360 --> 00:36:42,560
DEMONSTRATION OF THIS SORT OF

1104
00:36:42,560 --> 00:36:44,080
SCAFFOLDING, WE CAN TAKE A METAL

1105
00:36:44,080 --> 00:36:47,400
BINDING SITE LIKE THIS SQUARE

1106
00:36:47,400 --> 00:36:51,120
PLANAR ZINC SITE, AND WE CAN PUT

1107
00:36:51,120 --> 00:36:52,320
BASICALLY ADD THE SAME NOISE

1108
00:36:52,320 --> 00:36:54,600
DISTRIBUTION TO EACH OF --

1109
00:36:54,600 --> 00:36:56,120
BASICALLY AROUND IT, THAT'S SORT

1110
00:36:56,120 --> 00:36:57,280
OF SHOWN HERE, AND THEN BY GOING

1111
00:36:57,280 --> 00:36:59,120
THROUGH THIS DENOISING

1112
00:36:59,120 --> 00:37:00,640
TRAJECTORY, AS YOU CAN SEE HERE,

1113
00:37:00,640 --> 00:37:05,320
WE ULTIMATELY END UP WITH POE

1114
00:37:05,320 --> 00:37:06,240
PROTEINS THAT ASSEMBLE INTO

1115
00:37:06,240 --> 00:37:07,760
SOMETHING WITH -- IT'S A

1116
00:37:07,760 --> 00:37:12,200
TETRAMER AND IT HOLDS THESE IN

1117
00:37:12,200 --> 00:37:13,480
JUST THE RIGHT ARRANGEMENT.

1118
00:37:13,480 --> 00:37:14,920
THESE PROTEINS BIND METAL VERY

1119
00:37:14,920 --> 00:37:15,240
TIGHTLY.

1120
00:37:15,240 --> 00:37:20,960
THIS IS USING ITC, ISOTHERMAL

1121
00:37:20,960 --> 00:37:22,240
TITRATION -- WE CAN'T ACTUALLY

1122
00:37:22,240 --> 00:37:23,560
MEASURE HOW TIGHT THAT IS, IT'S

1123
00:37:23,560 --> 00:37:25,040
TOO TIGHT, AND THEY HAVE VERY

1124
00:37:25,040 --> 00:37:26,760
CLOSE TO THE DESIGN STRUCTURE AS

1125
00:37:26,760 --> 00:37:28,520
YOU CAN SEE HERE.

1126
00:37:28,520 --> 00:37:32,040
FROM THE EM.

1127
00:37:32,040 --> 00:37:33,880
NOW, COMING BACK TO THE BINDER

1128
00:37:33,880 --> 00:37:35,400
DESIGN PROBLEM, WHICH I TALKED

1129
00:37:35,400 --> 00:37:36,720
ABOUT IN THE BEGINNING, SO HERE

1130
00:37:36,720 --> 00:37:39,160
WHAT WE DO IS WE JUST START BY

1131
00:37:39,160 --> 00:37:41,200
PICKING A TARGET LIKE, YOU KNOW,

1132
00:37:41,200 --> 00:37:42,640
CORONAVIRUS OR WHATEVER, AND A

1133
00:37:42,640 --> 00:37:44,320
FEW RESIDUES ON THE TARGET, AND

1134
00:37:44,320 --> 00:37:46,720
THEN WE PLACE OUR RANDOM NOISE

1135
00:37:46,720 --> 00:37:48,360
CLOUD AROUND THOSE RESIDUES, AND

1136
00:37:48,360 --> 00:37:50,320
GO THROUGH THIS DENOISING

1137
00:37:50,320 --> 00:37:51,880
PROCESS, AND NOW WE HAVE A

1138
00:37:51,880 --> 00:37:55,480
PROTEIN THAT'S RIGHT ABOVE

1139
00:37:55,480 --> 00:37:57,880
THE -- THAT'S RIGHT ABOVE THAT

1140
00:37:57,880 --> 00:38:01,400
SITE, AND WE HAVE DONE THAT NOW

1141
00:38:01,400 --> 00:38:02,840
FOR A NUMBER OF TARGETS.

1142
00:38:02,840 --> 00:38:05,760
THIS ACTUALLY TOOK ABOUT A WEEK

1143
00:38:05,760 --> 00:38:07,760
IN TOTAL, INCLUDING THE

1144
00:38:07,760 --> 00:38:08,520
EXPERIMENTS.

1145
00:38:08,520 --> 00:38:09,720
WE WERE TRYING TO SEE HOW FAST

1146
00:38:09,720 --> 00:38:11,160
WE COULD DO THIS.

1147
00:38:11,160 --> 00:38:14,160
SO THE AFFINITIES ARE NOT GREAT.

1148
00:38:14,160 --> 00:38:16,200
THERE'S STILL ROOM FOR

1149
00:38:16,200 --> 00:38:17,120
IMPROVEMENT WITH THIS METHOD,

1150
00:38:17,120 --> 00:38:18,640
BUT IMPORTANTLY, WE WERE ABLE TO

1151
00:38:18,640 --> 00:38:20,200
MAKE BINDERS FROM SCRATCH

1152
00:38:20,200 --> 00:38:21,800
WITHOUT TESTING A LARGE NUMBER

1153
00:38:21,800 --> 00:38:22,360
OF DESIGNS.

1154
00:38:22,360 --> 00:38:24,240
SO BEFORE WE WOULD HAVE TO TEST

1155
00:38:24,240 --> 00:38:25,880
TENS OF THOUSANDS OF DESIGNS TO

1156
00:38:25,880 --> 00:38:27,280
FIND ONES THAT BOUND, AND HERE

1157
00:38:27,280 --> 00:38:29,640
WE'RE JUST TESTING DOZENS AND

1158
00:38:29,640 --> 00:38:32,000
GETTING BINDERS IN THE NANOMOLAR

1159
00:38:32,000 --> 00:38:32,200
RANGE.

1160
00:38:32,200 --> 00:38:35,080
SO IN TERMS OF THE PANDEMIC

1161
00:38:35,080 --> 00:38:36,000
RESPONSE, I THINK WE'RE GETTING

1162
00:38:36,000 --> 00:38:37,640
CLOSE TO THE POINT THAT IF A NEW

1163
00:38:37,640 --> 00:38:38,920
VIRUS DID BREAK OUT AND WE HAD

1164
00:38:38,920 --> 00:38:40,400
THE SEQUENCE, WE COULD PREDICT

1165
00:38:40,400 --> 00:38:42,040
THE STRUCTURE AND THEN USE THESE

1166
00:38:42,040 --> 00:38:44,440
METHODS TO VERY RAPIDLY GENERATE

1167
00:38:44,440 --> 00:38:44,680
BINDERS.

1168
00:38:44,680 --> 00:38:47,320
THE WORK WE HAVE TO DO IS TO GET

1169
00:38:47,320 --> 00:38:53,360
THESE TO BE PICO MO PICOMOLAR IF

1170
00:38:53,360 --> 00:38:53,600
NANOMOLAR.

1171
00:38:53,600 --> 00:38:55,600
THEN AS MY FINAL SLIDE, THIS

1172
00:38:55,600 --> 00:38:57,600
DEMONSTRATION THAT THIS ISN'T

1173
00:38:57,600 --> 00:39:00,120
CRAZY, THIS IS TAKING

1174
00:39:00,120 --> 00:39:04,640
PARATHYROID HORMONE AND NOW

1175
00:39:04,640 --> 00:39:06,800
BASICALLY PUTTING RANDOM

1176
00:39:06,800 --> 00:39:09,000
DISTRIBUTION AROUND IT, AND THEN

1177
00:39:09,000 --> 00:39:10,240
RUNNING THE DENOISING PROCESS

1178
00:39:10,240 --> 00:39:12,920
AND YOU CAN AGAIN SEE A PROTEIN

1179
00:39:12,920 --> 00:39:17,200
ASSEMBLING AROUND THE

1180
00:39:17,200 --> 00:39:17,720
PARATHYROID HORMONE.

1181
00:39:17,720 --> 00:39:18,760
WHEN WE MAKE THESE PROTEINS IN

1182
00:39:18,760 --> 00:39:20,560
THE LAB, THEY BIND EXTREMELY

1183
00:39:20,560 --> 00:39:22,640
TIGHTLY, SO WE'VE DONE THIS FOR

1184
00:39:22,640 --> 00:39:23,840
PARATHYROID HORMONE HERE AND

1185
00:39:23,840 --> 00:39:26,160
ANOTHER HELICAL PROTEIN HERE AND

1186
00:39:26,160 --> 00:39:28,200
YOU CAN SEE THERE'S ESSENTIALLY

1187
00:39:28,200 --> 00:39:31,520
NO DISSOCIATION IN THIS

1188
00:39:31,520 --> 00:39:31,960
EXPERIMENT.

1189
00:39:31,960 --> 00:39:33,920
SO WE'RE GETTING SUBNANOMOLAR

1190
00:39:33,920 --> 00:39:34,920
BINDING STRAIGHT OUT OF THE

1191
00:39:34,920 --> 00:39:38,760
COMPUTE E WCOMPUTER, WHICH IS SE

1192
00:39:38,760 --> 00:39:39,840
AND OTHERS HAVE NEVER REALLY

1193
00:39:39,840 --> 00:39:41,160
SEEN BEFORE FOR ANY KIND OF

1194
00:39:41,160 --> 00:39:41,880
BINDING DESIGN THING.

1195
00:39:41,880 --> 00:39:43,480
I THINK THIS IS AN EASY CASE

1196
00:39:43,480 --> 00:39:44,880
BECAUSE YOU CAN SEE THE PROTEIN

1197
00:39:44,880 --> 00:39:46,560
BASICALLY JUST HAS TO CRADLE

1198
00:39:46,560 --> 00:39:47,200
THIS LONG HELIX.

1199
00:39:47,200 --> 00:39:49,920
BUT I THINK THERE'S HOPE NOW FOR

1200
00:39:49,920 --> 00:39:52,800
GENERAL TARGETS AND GETTING

1201
00:39:52,800 --> 00:39:54,240
PICOMOLAR BINDERS REALLY QUITE

1202
00:39:54,240 --> 00:39:54,560
FAST.

1203
00:39:54,560 --> 00:39:55,760
SO THAT'S WHAT I WANTED TO TELL

1204
00:39:55,760 --> 00:39:57,840
YOU ABOUT TODAY.

1205
00:39:57,840 --> 00:40:00,480
I'VE HAD MANY ABSOLUTELY AMAZING

1206
00:40:00,480 --> 00:40:01,880
COLLABORATORS ON THIS.

1207
00:40:01,880 --> 00:40:08,240
SO THE ANTIVIRAL DESIGN WORK WAS

1208
00:40:08,240 --> 00:40:10,240
DONE WITH MANY, MANY

1209
00:40:10,240 --> 00:40:11,640
COLLABORATORS IN MANY DIFFERENT

1210
00:40:11,640 --> 00:40:13,640
PLACES FOR ALL THE ANIMAL

1211
00:40:13,640 --> 00:40:14,640
EXPERIMENTS.

1212
00:40:14,640 --> 00:40:23,000
MIKE DIAMOND'S LAB, WE DID A LOT

1213
00:40:23,000 --> 00:40:25,120
OF THE ANIMAL EXPERIMENTS WITH

1214
00:40:25,120 --> 00:40:26,560
THE COVID BINDER.

1215
00:40:26,560 --> 00:40:29,960
THE AMYLOID PEPTIDE BINDERS WERE

1216
00:40:29,960 --> 00:40:35,760
DESIGNED BY DANNY SATO.

1217
00:40:35,760 --> 00:40:38,360
THE ANTIBODY CAGES FOR DELIVERY

1218
00:40:38,360 --> 00:40:41,040
WERE DESIGNED BY ERIN AND ROBBY

1219
00:40:41,040 --> 00:40:41,800
AND ANDREW.

1220
00:40:41,800 --> 00:40:43,200
THE PROTEIN NANOPORES, WHAT

1221
00:40:43,200 --> 00:40:45,520
WE'RE HOPING FOR THE NEXT

1222
00:40:45,520 --> 00:40:46,680
GENERATION PROTEIN SEQUENCING,

1223
00:40:46,680 --> 00:40:54,280
FOR EXAMPLE, WERE DESIGNED BY

1224
00:40:54,280 --> 00:41:00,760
SAMUEL, SAGARDIP, ANASTASIA AND

1225
00:41:00,760 --> 00:41:02,560
CAROLIN.

1226
00:41:02,560 --> 00:41:03,680
THE PROTEIN HALLUCINATION

1227
00:41:03,680 --> 00:41:10,400
APPROACH I DESCRIENED WAS

1228
00:41:10,400 --> 00:41:13,480
DEVELOPED BY ALEXIS, BASILE AND

1229
00:41:13,480 --> 00:41:15,560
LUK AS.

1230
00:41:15,560 --> 00:41:18,480
ANDY AND CHRIS DESIGNED THE

1231
00:41:18,480 --> 00:41:19,800
LUCIFERASE, AND ROSETTAFOLD IS

1232
00:41:19,800 --> 00:41:27,200
THE WORK OF MINKYUNG AND FRANK.

1233
00:41:27,200 --> 00:41:29,720
JOE AND DAVID REALLY LED THE

1234
00:41:29,720 --> 00:41:30,920
EFFORT WITH IMPORTANT

1235
00:41:30,920 --> 00:41:33,000
CONTRIBUTIONS FROM EVERYONE

1236
00:41:33,000 --> 00:41:37,720
LISTED HERE.

1237
00:41:37,720 --> 00:41:39,760
THERE ARE A LOT OF PEOPLE WHO I

1238
00:41:39,760 --> 00:41:41,760
NEED TO THANK WHO AREN'T ON THIS

1239
00:41:41,760 --> 00:41:42,880
SLIDE, BUT IN PARTICULAR, MY

1240
00:41:42,880 --> 00:41:44,600
COLLEAGUES AT THE INSTITUTE FOR

1241
00:41:44,600 --> 00:41:45,840
PROTEIN DESIGN WHO REALLY MAKE A

1242
00:41:45,840 --> 00:41:47,840
LOT OF THINGS HAPPEN.

1243
00:41:47,840 --> 00:41:48,840
SO THANKS FOR YOUR ATTENTION.

1244
00:41:48,840 --> 00:41:49,880
I'D BE HAPPY TO TAKE ANY

1245
00:41:49,880 --> 00:41:50,160
QUESTIONS.

1246
00:41:50,160 --> 00:41:56,880
[APPLAUSE]

1247
00:41:56,880 --> 00:41:58,400
>>THANK YOU.

1248
00:41:58,400 --> 00:42:00,640
ANYONE IS FREE TO USE THE

1249
00:42:00,640 --> 00:42:00,920
MICROPHONES.

1250
00:42:00,920 --> 00:42:02,120
WE DO HAVE MANY QUESTIONS

1251
00:42:02,120 --> 00:42:03,680
ONLINE, SO LET'S START THERE.

1252
00:42:03,680 --> 00:42:06,080
PERHAPS YOU KNOW THIS FELLOW.

1253
00:42:06,080 --> 00:42:09,040
OMAR HABIB OVER IN OXFORD.

1254
00:42:09,040 --> 00:42:10,040
GREAT TALK, DAVID.

1255
00:42:10,040 --> 00:42:13,560
DOES YOUR APPROACH TAKE PTNs

1256
00:42:13,560 --> 00:42:14,760
INTO CONTRIBUTION WHEN

1257
00:42:14,760 --> 00:42:15,280
PREDICTING STRUCTURE?

1258
00:42:15,280 --> 00:42:17,840
>>YEAH, SO WE ARE JUST
WORKING

1259
00:42:17,840 --> 00:42:18,520
ON THAT NOW.

1260
00:42:18,520 --> 00:42:20,240
SO THE ORIGINAL VERSION OF

1261
00:42:20,240 --> 00:42:22,640
ROSETTAFOLD, LIKE ALPHAFOLD, DID

1262
00:42:22,640 --> 00:42:24,320
NOT INCLUDE POST TRANSLATIONAL

1263
00:42:24,320 --> 00:42:26,760
MODIFICATIONS.

1264
00:42:26,760 --> 00:42:27,920
BUT WE'RE WORKING ON A VERSION

1265
00:42:27,920 --> 00:42:30,240
NOW THAT WILL DO THAT.

1266
00:42:30,240 --> 00:42:31,600
>>OKAY.

1267
00:42:31,600 --> 00:42:33,440
I'M KNOLL GIST QUESTION.

1268
00:42:33,440 --> 00:42:35,520
SINCE THE SYNTHETIC MINI BINDERS

1269
00:42:35,520 --> 00:42:37,320
ARE FOREIGN, THE PATIENT'S

1270
00:42:37,320 --> 00:42:38,920
IMMUNE SYSTEM CAN DEVELOP

1271
00:42:38,920 --> 00:42:39,760
NEUTRALIZING ANTIBODIES AGAINST

1272
00:42:39,760 --> 00:42:40,560
THEM.

1273
00:42:40,560 --> 00:42:42,760
WOUWOULDN'T THAT LIMIT THEIR LOG

1274
00:42:42,760 --> 00:42:43,400
TERM USE?

1275
00:42:43,400 --> 00:42:44,520
>>YES AND THAT WAS THE
BIG

1276
00:42:44,520 --> 00:42:45,720
THING WE WERE WORRIED ABOUT WHEN

1277
00:42:45,720 --> 00:42:48,160
WE STARTED PUTTING THESE DESIGN

1278
00:42:48,160 --> 00:42:49,240
PROTEINS, WHICH ARE FOREIGN,

1279
00:42:49,240 --> 00:42:50,000
INTO ANIMALS.

1280
00:42:50,000 --> 00:42:51,840
AND ULTIMATELY IN PEOPLE.

1281
00:42:51,840 --> 00:42:55,400
BUT THE RESULTS ARE ACTUALLY

1282
00:42:55,400 --> 00:42:57,680
PRETTY CLEAR.

1283
00:42:57,680 --> 00:43:00,080
SO THE SMALL PROTEINS DO NOT

1284
00:43:00,080 --> 00:43:02,640
ELICIT STRONG IMMUNE RESPONSES,

1285
00:43:02,640 --> 00:43:05,680
SO FOR BOTH THE -- BOTH IN THE

1286
00:43:05,680 --> 00:43:11,920
CASE OF INFLUENZA AND THE

1287
00:43:11,920 --> 00:43:13,800
CORONAVIRUS, I DESCRIBED

1288
00:43:13,800 --> 00:43:15,480
THESE -- I TOLD YOU THEY HAD

1289
00:43:15,480 --> 00:43:17,960
STRONG -- THEY HAD POTENT

1290
00:43:17,960 --> 00:43:19,200
THERAPEUTIC AND PROPHYLACTIC

1291
00:43:19,200 --> 00:43:19,520
EFFECT.

1292
00:43:19,520 --> 00:43:20,800
YOU COULD ACTUALLY ADMINISTER

1293
00:43:20,800 --> 00:43:22,840
THEM OVER A THREE-WEEK PERIOD

1294
00:43:22,840 --> 00:43:24,360
EVERY THREE DAYS TO TRY AND

1295
00:43:24,360 --> 00:43:25,000
ELICIT IMMUNE RESPONSE.

1296
00:43:25,000 --> 00:43:26,560
AND THERE'S NO DECREASE IN

1297
00:43:26,560 --> 00:43:27,560
EFFICACY.

1298
00:43:27,560 --> 00:43:29,520
SO NEUTRALIZING ANTIBODIES DON'T

1299
00:43:29,520 --> 00:43:31,280
SEEM TO GET MADE, AND WHEN WE

1300
00:43:31,280 --> 00:43:32,760
LOOK JUST FOR ANTIBODY

1301
00:43:32,760 --> 00:43:33,640
RESPONSES, THEY'RE PRETTY WEAK.

1302
00:43:33,640 --> 00:43:35,160
AND I THINK THAT'S BECAUSE THE

1303
00:43:35,160 --> 00:43:36,680
PROTEINS ARE VERY STABLE,

1304
00:43:36,680 --> 00:43:38,400
THEY'RE SMALL, THEY'RE VERY

1305
00:43:38,400 --> 00:43:39,600
SOLUBLE, SO THEY'RE PROBABLY NOT

1306
00:43:39,600 --> 00:43:41,680
TAKEN UP WELL BY DENDRITIC CELLS

1307
00:43:41,680 --> 00:43:42,960
AND THEY'RE VERY STABLE SO

1308
00:43:42,960 --> 00:43:44,040
THEY'RE PROBABLY NOT PRESENTED

1309
00:43:44,040 --> 00:43:45,120
THAT WELL.

1310
00:43:45,120 --> 00:43:47,360
>>OKAY.

1311
00:43:47,360 --> 00:43:49,160
FROM ROLAND OWENS, WOULD AN

1312
00:43:49,160 --> 00:43:52,240
ARTIFICIAL PROTEIN CAPABLE OF

1313
00:43:52,240 --> 00:43:54,120
INDUCING TRANSDIFFERENTIATION

1314
00:43:54,120 --> 00:43:58,560
ALSO BE A POTENTIAL TERATOGEN?

1315
00:43:58,560 --> 00:44:00,560
>>YEAH, SO I THINK FOR
THESE

1316
00:44:00,560 --> 00:44:01,640
NOVEL AGONISTS, WHERE WE'RE

1317
00:44:01,640 --> 00:44:04,160
TRYING TO GET, SAY, SULFATE

1318
00:44:04,160 --> 00:44:05,040
CHANGES, OBVIOUSLY THEY WOULD

1319
00:44:05,040 --> 00:44:06,400
HAVE TO BE TARGETED VERY

1320
00:44:06,400 --> 00:44:06,680
CAREFULLY.

1321
00:44:06,680 --> 00:44:11,640
I SHOULD SAY, WE ALSO HAVE A DE

1322
00:44:11,640 --> 00:44:14,600
NOVO DESIGNED CYTOKINE MIMIC

1323
00:44:14,600 --> 00:44:15,280
THAT'S IN CLINICAL TRIALS AND

1324
00:44:15,280 --> 00:44:17,560
THAT HAS NOT ELICITED ANY

1325
00:44:17,560 --> 00:44:18,400
SIGNIFICANT IMMUNE RESPONSE

1326
00:44:18,400 --> 00:44:19,120
COMING BACK TO THE PREVIOUS

1327
00:44:19,120 --> 00:44:20,360
QUESTION.

1328
00:44:20,360 --> 00:44:22,000
>>YOU MENTIONED THAT

1329
00:44:22,000 --> 00:44:25,800
ROSETTAFOLD CAN NOW PREDICT

1330
00:44:25,800 --> 00:44:26,960
BINDERS TO NUCLEIC ACIDS.

1331
00:44:26,960 --> 00:44:28,600
CAN YOU PREDICT WHAT SEQUENCE

1332
00:44:28,600 --> 00:44:30,000
PROTEIN MIGHT BE ABLE TO BIND,

1333
00:44:30,000 --> 00:44:32,640
AND RELATED QUESTION, CAN YOU DO

1334
00:44:32,640 --> 00:44:33,560
SIMILAR THINGS FOR SMALL

1335
00:44:33,560 --> 00:44:36,960
MOLECULES TO SAY WHAT ENZYME

1336
00:44:36,960 --> 00:44:38,400
MIGHT ACT ON THAT SORT OF THING?

1337
00:44:38,400 --> 00:44:39,840
>>YEAH, WELL --

1338
00:44:39,840 --> 00:44:41,040
>>I'M SORRY, CAN YOU SUM
UP

1339
00:44:41,040 --> 00:44:41,480
THAT QUESTION?

1340
00:44:41,480 --> 00:44:42,800
I DON'T KNOW IF IT WAS HEARD, IF

1341
00:44:42,800 --> 00:44:43,520
THE MIC WAS ON.

1342
00:44:43,520 --> 00:44:43,840
>>RIGHT.

1343
00:44:43,840 --> 00:44:47,400
SO THE QUESTION IS, I SHOWED YOU

1344
00:44:47,400 --> 00:44:51,040
THAT ROSETTAFOLD CAN NOW PREDICT

1345
00:44:51,040 --> 00:44:52,360
PROTEIN DNA COMPLEXES AND THE

1346
00:44:52,360 --> 00:44:54,160
QUESTION IS CAN IT PREDICT WHAT

1347
00:44:54,160 --> 00:44:56,360
DNA SEQUENCE A PROTEIN WILL

1348
00:44:56,360 --> 00:44:59,080
BIEND TO AND SIM LAICALLY FOR

1349
00:44:59,080 --> 00:45:02,200
SMALL MOLECULES, COULD YOU

1350
00:45:02,200 --> 00:45:03,160
PREDICT WHAT SMALL MOLECULE IT

1351
00:45:03,160 --> 00:45:04,040
WAS GOING TO BIND TO.

1352
00:45:04,040 --> 00:45:05,800
I WOULD SAY AT THIS STAGE, IF

1353
00:45:05,800 --> 00:45:08,440
YOU GIVE ROSETTAFOLD A PIECE OF

1354
00:45:08,440 --> 00:45:09,840
DNA THAT'S NOT TOO BIG, IT WILL

1355
00:45:09,840 --> 00:45:10,960
CORRECTLY BIND TO THE RIGHT

1356
00:45:10,960 --> 00:45:12,840
SITE, BUT I THINK THE ABILITY TO

1357
00:45:12,840 --> 00:45:13,840
DISCRIMINATE AMONG ALL POSSIBLE

1358
00:45:13,840 --> 00:45:15,480
SITES TO PREDICT WHAT THE RIGHT

1359
00:45:15,480 --> 00:45:17,280
ONE IS IS NOT SO GOOD, BUT WE'RE

1360
00:45:17,280 --> 00:45:19,560
ACTUALLY NOW JUST STARTING TO

1361
00:45:19,560 --> 00:45:20,320
EXPLICITLY TRAIN IT BECAUSE

1362
00:45:20,320 --> 00:45:22,880
THERE'S A LOT OF EXPERIMENTAL

1363
00:45:22,880 --> 00:45:24,280
SPECIFICITY DATA.

1364
00:45:24,280 --> 00:45:25,800
THESE DEEP LEARNING MODELS ARE

1365
00:45:25,800 --> 00:45:27,240
ONLY AS GOOD AS THE TRAINING

1366
00:45:27,240 --> 00:45:28,360
SETS YOU GIVE THEM SO WE THINK

1367
00:45:28,360 --> 00:45:29,560
WE CAN IMPROVE THAT.

1368
00:45:29,560 --> 00:45:31,640
AND THEN WE AND OTHERS ARE

1369
00:45:31,640 --> 00:45:33,360
DEVELOPING ROSETTAFOLD FOR SMALL

1370
00:45:33,360 --> 00:45:34,440
MOLECULE BINDING, AND THAT'S

1371
00:45:34,440 --> 00:45:37,480
LOOKING PROMISING.

1372
00:45:37,480 --> 00:45:40,000
AGAIN, I THINK IT'S EASIER TO

1373
00:45:40,000 --> 00:45:42,320
PREDICT THE POSE THAN TO PREDICT

1374
00:45:42,320 --> 00:45:44,200
SPECIFICITY BUT I THINK THAT

1375
00:45:44,200 --> 00:45:44,640
WILL ALL COME.

1376
00:45:44,640 --> 00:45:46,280
>>A FEW MORE QUESTIONS
AND

1377
00:45:46,280 --> 00:45:47,400
REMINDER WE'RE GOING TO HAVE OUR

1378
00:45:47,400 --> 00:45:48,400
RECEPTION, THE FIRST ONE IN MORE

1379
00:45:48,400 --> 00:45:50,680
THAN TWO YEARS, JUST OUTSIDE

1380
00:45:50,680 --> 00:45:53,920
THIS AUDITORIUM.

1381
00:45:53,920 --> 00:45:55,040
OKAY.

1382
00:45:55,040 --> 00:45:57,880
QUESTION FROM MATT PEDELTON.

1383
00:45:57,880 --> 00:45:58,880
ANY CONSIDERATION FOR ATTACHING

1384
00:45:58,880 --> 00:46:01,760
AN FC DOMAIN ON TO THESE VIRAL

1385
00:46:01,760 --> 00:46:03,360
INHIBITORS IN ORDER TO INHIBIT

1386
00:46:03,360 --> 00:46:05,360
AND ELICIT AN APPROPRIATE

1387
00:46:05,360 --> 00:46:07,440
ADAPTIVE IMMUNE EFFECTOR

1388
00:46:07,440 --> 00:46:07,720
FUNCTION?

1389
00:46:07,720 --> 00:46:09,200
>>SO BRETT CASE AND MIKE

1390
00:46:09,200 --> 00:46:11,800
DIAMOND DID THAT, AND WITH THE

1391
00:46:11,800 --> 00:46:13,360
CORONAVIRUS BINDERS, AND THEY'RE

1392
00:46:13,360 --> 00:46:15,400
VERY -- THEY WORK VERY WELL.

1393
00:46:15,400 --> 00:46:16,680
IT'S NOT TOTALLY CLEAR IN THAT

1394
00:46:16,680 --> 00:46:18,240
CASE WHAT THE ADVANTAGE OVER

1395
00:46:18,240 --> 00:46:19,080
ANTIBODIES ARE.

1396
00:46:19,080 --> 00:46:22,080
THE ONE CASE COULD BE THAT IF WE

1397
00:46:22,080 --> 00:46:23,280
CAN DESIGN BINDERS TO BIND TO

1398
00:46:23,280 --> 00:46:24,600
SITES THAT AREN'T ACCESSIBLE

1399
00:46:24,600 --> 00:46:25,800
THAT ANTIBODIES DON'T RECOGNIZE.

1400
00:46:25,800 --> 00:46:28,320
BUT YOU CAN PUT FCs ON THEM,

1401
00:46:28,320 --> 00:46:30,720
YOU EXTEND THE HALF-LIFE.

1402
00:46:30,720 --> 00:46:32,440
DEB FULLER'S LAB HAS DONE THAT

1403
00:46:32,440 --> 00:46:33,600
WITH THE FLU AS WELL, SO THAT

1404
00:46:33,600 --> 00:46:34,480
CAN CERTAINLY BE DONE.

1405
00:46:34,480 --> 00:46:35,280
>>OKAY.

1406
00:46:35,280 --> 00:46:38,240
AND MATT FOLLOWS UP.

1407
00:46:38,240 --> 00:46:39,600
SO, THIS DENOISING PROCESS IS

1408
00:46:39,600 --> 00:46:42,080
NOT STRUCTURALLY CONSTRAINED THE

1409
00:46:42,080 --> 00:46:47,280
SAME AS BACKBONE/SIDE CHAIN

1410
00:46:47,280 --> 00:46:50,560
PERTURBATION APPROACH, THE WORDS

1411
00:46:50,560 --> 00:46:53,520
DEPARTMENT DIDN'T COME THROUGH E

1412
00:46:53,520 --> 00:46:55,840
ORIGINAL RO ROSETTA WHERE EACH P

1413
00:46:55,840 --> 00:46:57,680
WAS RELATED TO PREVIOUS STEPS BY

1414
00:46:57,680 --> 00:46:59,280
SPECIFIC CHANGES IN BACKBONE

1415
00:46:59,280 --> 00:46:59,720
ANGLES?

1416
00:46:59,720 --> 00:47:00,480
COULD YOU FOLLOW THAT?

1417
00:47:00,480 --> 00:47:01,040
>>RIGHT.

1418
00:47:01,040 --> 00:47:03,000
SO BASICALLY THE WAY -- SO THIS

1419
00:47:03,000 --> 00:47:05,760
PROBLEM OF HOW DO YOU SAMPLE

1420
00:47:05,760 --> 00:47:06,880
THROUGH POSSIBLE BACKBONES TO

1421
00:47:06,880 --> 00:47:08,760
MAKE UP NEW STRUCTURES IS ONE

1422
00:47:08,760 --> 00:47:10,160
THAT WE WORKED ON FOR A LONG

1423
00:47:10,160 --> 00:47:12,160
TIME IN THE CONTEXT OF THE

1424
00:47:12,160 --> 00:47:13,000
ROSETTA PHYSICAL MODEL.

1425
00:47:13,000 --> 00:47:14,800
IN THAT CASE, YOU KEEP THE CHAIN

1426
00:47:14,800 --> 00:47:16,320
CONNECTED AND ALL THE BOND LINKS

1427
00:47:16,320 --> 00:47:17,960
AND BOND ANGLES ARE IDEAL, AND

1428
00:47:17,960 --> 00:47:19,440
YOU'RE JUST SAMPLING THE

1429
00:47:19,440 --> 00:47:21,040
BACKBONE TORSION ANGLES.

1430
00:47:21,040 --> 00:47:22,840
THAT'S HOW YOU SEARCH AROUND IN

1431
00:47:22,840 --> 00:47:23,560
SPACE.

1432
00:47:23,560 --> 00:47:26,040
IN THIS DIFFUSION APPROACH, IT'S

1433
00:47:26,040 --> 00:47:26,560
NOT LIKE THAT.

1434
00:47:26,560 --> 00:47:30,160
THE RESIDUES ARE ALL SEPARATE.

1435
00:47:30,160 --> 00:47:31,280
BASICALLY EACH RESIDUE HAS A

1436
00:47:31,280 --> 00:47:33,280
POSITION AND AN ORIENTATION, AND

1437
00:47:33,280 --> 00:47:35,920
WHAT YOU'RE DIFFUSING OVER, WHAT

1438
00:47:35,920 --> 00:47:37,440
YOU'RE SAMPLING ARE THOSE

1439
00:47:37,440 --> 00:47:42,080
POSITIONS AND ORIENTATIONS.

1440
00:47:42,080 --> 00:47:43,360
AND SO IT'S REALLY QUITE

1441
00:47:43,360 --> 00:47:44,560
DIFFERENT IN THAT WAY.

1442
00:47:44,560 --> 00:47:46,360
WE'VE TRIED TO PUT SOME OF THE

1443
00:47:46,360 --> 00:47:49,560
PHYSICS THAT'S IN ROSETTA INTO

1444
00:47:49,560 --> 00:47:50,560
ROSETTAFOLD BUT IT REALLY HASN'T

1445
00:47:50,560 --> 00:47:51,400
HELPED THAT MUCH AND I THINK

1446
00:47:51,400 --> 00:47:52,880
THAT'S SORT OF A TESTAMENT TO

1447
00:47:52,880 --> 00:47:54,680
HOW RICH THE PROTEIN STRUCTURE

1448
00:47:54,680 --> 00:47:55,280
DATABASE IS.

1449
00:47:55,280 --> 00:47:56,600
THERE'S SO MUCH INFORMATION IN

1450
00:47:56,600 --> 00:47:58,040
THERE, THE NETWORK KIND OF

1451
00:47:58,040 --> 00:47:59,480
LEARNS ALL THE RULES ABOUT

1452
00:47:59,480 --> 00:48:00,360
CONNECTIVITY WITHOUT HAVING TO

1453
00:48:00,360 --> 00:48:01,880
BE PROVIDED WITH ADDITIONAL

1454
00:48:01,880 --> 00:48:03,560
INFORMATION.

1455
00:48:03,560 --> 00:48:04,120
>>OKAY.

1456
00:48:04,120 --> 00:48:05,600
HAVE YOU CONSIDERED EXPANDING

1457
00:48:05,600 --> 00:48:08,720
BEYOND THE CHEMISTRY OF THE

1458
00:48:08,720 --> 00:48:10,040
CANONICAL AMINO ACIDS?

1459
00:48:10,040 --> 00:48:11,520
>>YEAH, SO THAT'S AN
IMPORTANT

1460
00:48:11,520 --> 00:48:12,360
DIFFERENCE BETWEEN THE

1461
00:48:12,360 --> 00:48:13,160
PHYSICALLY BASED APPROACH AND

1462
00:48:13,160 --> 00:48:14,360
THE DEEP LEARNING APPROACH.

1463
00:48:14,360 --> 00:48:15,760
WITH THE PHYSICALLY BASED

1464
00:48:15,760 --> 00:48:17,880
APPROACH, WE'VE DONE THAT, AND

1465
00:48:17,880 --> 00:48:20,280
SO WE CAN -- BUT THERE THE

1466
00:48:20,280 --> 00:48:21,400
PROBLEM IS MAKING THE PROTEINS,

1467
00:48:21,400 --> 00:48:22,960
BECAUSE IF YOU'RE PUTTING IN A

1468
00:48:22,960 --> 00:48:25,160
LOT OF UNNATURAL CHEMISTRIES, IT

1469
00:48:25,160 --> 00:48:26,680
GETS HARDER AND HARDER TO MAKE

1470
00:48:26,680 --> 00:48:27,360
THEM IN CELLS.

1471
00:48:27,360 --> 00:48:30,080
SO WE'VE REALLY FOCUSED ON SMALL

1472
00:48:30,080 --> 00:48:35,840
SICYCLIC PEPTIDES, WE'VE DESIGND

1473
00:48:35,840 --> 00:48:37,800
THEM SO THEY CAN GET ACROSS

1474
00:48:37,800 --> 00:48:39,520
MEMBRANES VERY READILY AND THEY

1475
00:48:39,520 --> 00:48:41,240
HAVE RIGID STRUCTURES SO NOW

1476
00:48:41,240 --> 00:48:43,720
WE'RE DEVELOPING METHODS TO MAKE

1477
00:48:43,720 --> 00:48:47,160
THEM BIND TIGHTLY TO -- OF

1478
00:48:47,160 --> 00:48:48,520
INTEREST BUT WE HAVE TO MAKE

1479
00:48:48,520 --> 00:48:49,440
THOSE CHEMICALLY.

1480
00:48:49,440 --> 00:48:52,320
>>FOR YOUR RF DIFFUSION
BINDING

1481
00:48:52,320 --> 00:48:54,200
FIBERS THAT YOU'RE ONLY GETTING

1482
00:48:54,200 --> 00:48:56,600
NANOMOLAR AND YOU NEED

1483
00:48:56,600 --> 00:48:58,600
PICOMOLAR, WOULD YOU DO -- CAN

1484
00:48:58,600 --> 00:49:02,080
YOU DO THAT IN SILICO OR WITH

1485
00:49:02,080 --> 00:49:03,080
A.I. NOWADAYS OR IS THAT

1486
00:49:03,080 --> 00:49:04,040
SOMETHING IN THE PIPELINE?

1487
00:49:04,040 --> 00:49:05,480
>>RIGHT, YOU'RE
ABSOLUTELY

1488
00:49:05,480 --> 00:49:06,400
RIGHT.

1489
00:49:06,400 --> 00:49:07,920
SO THE WAY THAT WE HAD BEEN

1490
00:49:07,920 --> 00:49:11,280
DOING THAT WITH THE ROSETTA

1491
00:49:11,280 --> 00:49:12,040
METHOD, LIKE THE CORONAVIRUS

1492
00:49:12,040 --> 00:49:13,920
BINDERS AND THE OTHER VIRUS

1493
00:49:13,920 --> 00:49:15,560
BINDERS I DESCRIBED AT THE TOP,

1494
00:49:15,560 --> 00:49:17,200
WE WOULD GET THE ORIGINAL HITS

1495
00:49:17,200 --> 00:49:18,840
FROM ROSETTA BUT THEN WE WOULD

1496
00:49:18,840 --> 00:49:20,000
BASICALLY BHAIK A LIBRARY WHERE

1497
00:49:20,000 --> 00:49:21,520
WE MUTATED EVERY RESIDUE TO

1498
00:49:21,520 --> 00:49:23,520
EVERY OTHER ONE AT A TIME AND

1499
00:49:23,520 --> 00:49:25,600
IDENTIFY SUBSTITUTIONS THAT

1500
00:49:25,600 --> 00:49:27,040
INCREASED AFFINITY AND WE WOULD

1501
00:49:27,040 --> 00:49:28,680
DEFINE THEM TO GET TO PICOMOLAR

1502
00:49:28,680 --> 00:49:29,400
LEVEL.

1503
00:49:29,400 --> 00:49:30,920
THAT'S SLOW, SO YOU'RE

1504
00:49:30,920 --> 00:49:33,560
ABSOLUTELY RIGHT, SO NOW OUR

1505
00:49:33,560 --> 00:49:35,840
CHALLENGES IS, AND I THINK IT'S

1506
00:49:35,840 --> 00:49:37,440
EMINENTLY SOLVABLE, IS DEVELOPED

1507
00:49:37,440 --> 00:49:39,240
SORT OF INSILICO WAYS OF DOING

1508
00:49:39,240 --> 00:49:42,160
THAT EXACTLY.

1509
00:49:42,160 --> 00:49:43,760
>>THERE'S A FEW MORE IF
YOU'RE

1510
00:49:43,760 --> 00:49:44,360
STILL GOING STRONG?

1511
00:49:44,360 --> 00:49:45,200
>>SURE.

1512
00:49:45,200 --> 00:49:48,640
>>THE CME CODE -- OH,
GOOD.

1513
00:49:48,640 --> 00:49:49,040
44435.

1514
00:49:49,040 --> 00:49:50,880
I WAS GOING TO RELY ON MY OWN

1515
00:49:50,880 --> 00:49:52,600
MEMORY, BUT IT GOT FLASHED UP

1516
00:49:52,600 --> 00:49:54,560
THERE.

1517
00:49:54,560 --> 00:49:55,480
OKAY.

1518
00:49:55,480 --> 00:49:56,360
WILLIAM ANDERSON ASKS, THANK YOU

1519
00:49:56,360 --> 00:49:57,680
FOR SHARING YOUR WORK.

1520
00:49:57,680 --> 00:49:59,080
THIS IS FASCINATING.

1521
00:49:59,080 --> 00:50:01,560
IN REGARDS TO THE DE NOVO

1522
00:50:01,560 --> 00:50:04,040
PROTEINS FOR

1523
00:50:04,040 --> 00:50:04,720
TREATMENT/THERAPIES, IS THERE --

1524
00:50:04,720 --> 00:50:06,160
IS THE GREATER CHALLENGE

1525
00:50:06,160 --> 00:50:08,480
DESIGNING EFFECTIVE PROTEINS OR

1526
00:50:08,480 --> 00:50:10,880
DELIVERING THOSE PROTEINS IN A

1527
00:50:10,880 --> 00:50:13,280
CORRECT WAY TO THE CORRECT

1528
00:50:13,280 --> 00:50:13,560
LOCATION?

1529
00:50:13,560 --> 00:50:14,560
>>THOSE ARE BOTH VERY
IMPORTANT

1530
00:50:14,560 --> 00:50:14,880
THINGS.

1531
00:50:14,880 --> 00:50:16,640
I WOULD SAY THE EVEN MORE

1532
00:50:16,640 --> 00:50:19,600
IMPORTANT THING IS SORT OF THE

1533
00:50:19,600 --> 00:50:22,880
BIOLOGY OR MEDICAL CONCEPT.

1534
00:50:22,880 --> 00:50:24,880
WE CAN NOW DESIGN PROTEINS THAT

1535
00:50:24,880 --> 00:50:25,400
BIND ALMOST ANYTHING.

1536
00:50:25,400 --> 00:50:26,920
THE QUESTION IS, HOW DO YOU CURE

1537
00:50:26,920 --> 00:50:27,520
DISEASE WITH THAT?

1538
00:50:27,520 --> 00:50:29,600
AND THE VIRUSES ARE PARTICULARLY

1539
00:50:29,600 --> 00:50:33,000
A CLEAR CASE, YOU JUST BLOCK THE

1540
00:50:33,000 --> 00:50:33,440
VIRUS.

1541
00:50:33,440 --> 00:50:34,920
AND THEN MAKING THESE NEW TYPES

1542
00:50:34,920 --> 00:50:37,320
OF AGONISTS, NEW TYPES OF

1543
00:50:37,320 --> 00:50:38,400
CYTOKINE OR GROWTH FACTOR

1544
00:50:38,400 --> 00:50:38,640
MIMICS.

1545
00:50:38,640 --> 00:50:40,560
SO I THINK YOU HAVE TO HAVE THE

1546
00:50:40,560 --> 00:50:41,320
BIOLOGICAL CONCEPT FOR HOW

1547
00:50:41,320 --> 00:50:44,240
YOU'RE GOING TO CURE DISEASE.

1548
00:50:44,240 --> 00:50:45,160
ANOTHER THING WE'VE BEEN LOOKING

1549
00:50:45,160 --> 00:50:46,920
AT IS MAKING PROTEINS THAT WILL

1550
00:50:46,920 --> 00:50:50,680
TARGET SPECIFIC PROTEINS FOR

1551
00:50:50,680 --> 00:50:51,080
DEGRADATION.

1552
00:50:51,080 --> 00:50:53,120
BUT BEYOND THAT, YOUR POINT IS

1553
00:50:53,120 --> 00:50:54,040
ABSOLUTELY RIGHT THAT NOT ONLY

1554
00:50:54,040 --> 00:50:55,440
DO YOU NEED BINDING BUT YOU NEED

1555
00:50:55,440 --> 00:50:56,640
THAT BINDING ACTIVITY AT THE

1556
00:50:56,640 --> 00:50:58,920
RIGHT PLACE AT THE RIGHT TIME.

1557
00:50:58,920 --> 00:51:00,400
SO WE'RE ALSO WORKING ON SORT OF

1558
00:51:00,400 --> 00:51:02,240
PROTEIN LOGIC SYSTEMS SO THAT

1559
00:51:02,240 --> 00:51:03,360
YOU ONLY HAVE THAT ACTIVITY IN

1560
00:51:03,360 --> 00:51:04,160
THE RIGHT PLACE.

1561
00:51:04,160 --> 00:51:07,640
>>OKAY.

1562
00:51:07,640 --> 00:51:11,160
AND RELATED, RISHI ASKS, CAN YOU

1563
00:51:11,160 --> 00:51:14,440
COMMENT ON THE IMMUNOGENICITY

1564
00:51:14,440 --> 00:51:16,040
AND CLEARANCE OF THESE DESIGN

1565
00:51:16,040 --> 00:51:16,480
PROTEINS IN VIVO?

1566
00:51:16,480 --> 00:51:19,040
>>SO AS I MENTIONED
EARLIER,

1567
00:51:19,040 --> 00:51:20,480
THE IMMUNOGENICITY OF THE DESIGN

1568
00:51:20,480 --> 00:51:21,680
PROTEINS IS LOWER THAN YOU WOULD

1569
00:51:21,680 --> 00:51:21,960
EXPECT.

1570
00:51:21,960 --> 00:51:23,520
SO THERE HAVE BEEN CLINICAL

1571
00:51:23,520 --> 00:51:26,080
TRIALS NOW OF OUR DESIGNED

1572
00:51:26,080 --> 00:51:27,920
CYTOKINE MIMIC, AND IT ELICITED

1573
00:51:27,920 --> 00:51:29,840
VERY LITTLE ANTIBODIES, AND ALL

1574
00:51:29,840 --> 00:51:31,560
THE ANTIVIRAL PROTEINS HAVE BEEN

1575
00:51:31,560 --> 00:51:33,000
IN MANY ANIMAL SYSTEMS AND

1576
00:51:33,000 --> 00:51:34,560
THERE'S NO SIGN OF ANY

1577
00:51:34,560 --> 00:51:36,880
NEUTRALIZING ANTIBODY RESPONSE.

1578
00:51:36,880 --> 00:51:38,120
CLEARANCE, THOUGH, CLEARANCE WE

1579
00:51:38,120 --> 00:51:38,880
UNDERSTAND VERY WELL SO THESE

1580
00:51:38,880 --> 00:51:40,640
ARE ALL SMALL PROTEIN SO THEY

1581
00:51:40,640 --> 00:51:43,640
GET FILTERED TOUT IN  OUT IN TH.

1582
00:51:43,640 --> 00:51:45,480
IF YOU DON'T DO ANYTHING ELSE.

1583
00:51:45,480 --> 00:51:46,840
AND THAT'S JUST A FUNCTION OF

1584
00:51:46,840 --> 00:51:47,240
SIZE.

1585
00:51:47,240 --> 00:51:48,640
SO FOR SOME APPLICATIONS, THAT'S

1586
00:51:48,640 --> 00:51:48,960
GOOD.

1587
00:51:48,960 --> 00:51:51,040
SO IF YOU WANT TO -- IF YOU'RE

1588
00:51:51,040 --> 00:51:52,120
DOING RADIONUCLIDE THERAPY, SO

1589
00:51:52,120 --> 00:51:55,240
YOU'RE ATTACHING A VERY POTENT

1590
00:51:55,240 --> 00:51:56,880
RADIO EMITTER TO THESE PROTEINS

1591
00:51:56,880 --> 00:51:59,000
AND TARGETING THEM TO A TUMOR,

1592
00:51:59,000 --> 00:51:59,760
FOR EXAMPLE, HAVING THINGS

1593
00:51:59,760 --> 00:52:01,160
WASHED OUT QUICKLY IS GOOD.

1594
00:52:01,160 --> 00:52:03,280
IF YOU WANT LONGER HALF LIVES,

1595
00:52:03,280 --> 00:52:04,840
THEN YOU HAVE TO EXTEND -- YOU

1596
00:52:04,840 --> 00:52:06,560
HAVE TO DO THINGS TO EXTEND THE

1597
00:52:06,560 --> 00:52:06,840
HALF-LIFE.

1598
00:52:06,840 --> 00:52:07,320
>>OKAY.

1599
00:52:07,320 --> 00:52:09,040
>>IF YOU PUT THEM IN --
THAT'S

1600
00:52:09,040 --> 00:52:11,320
THE NICE THING ABOUT THESE

1601
00:52:11,320 --> 00:52:12,560
RESPIRATORY ANTIVIRALS, YOU PUT

1602
00:52:12,560 --> 00:52:14,200
THEM INTO THE RESPIRATORY SYSTEM

1603
00:52:14,200 --> 00:52:15,120
AND THEY JUST STAY THERE.

1604
00:52:15,120 --> 00:52:16,640
SO LIKE I SAID, YOU CAN GET

1605
00:52:16,640 --> 00:52:17,800
PROTECTION FOUR DAYS IN ADVANCE.

1606
00:52:17,800 --> 00:52:18,400
>>OKAY.

1607
00:52:18,400 --> 00:52:19,960
LOOKS LIKE OUR NEXT PERSON IS

1608
00:52:19,960 --> 00:52:21,280
WATCHING THE WEDNESDAY EVENING

1609
00:52:21,280 --> 00:52:23,160
LECTURE SERIES FROM GERMANY.

1610
00:52:23,160 --> 00:52:27,040
I WAS VERY EXCITED TO SEE THE

1611
00:52:27,040 --> 00:52:29,440
PRE-PRINT FOR ROSETTAFOLD NA.

1612
00:52:29,440 --> 00:52:31,920
HOW FAR AWAY ARE YOU FROM MAKING

1613
00:52:31,920 --> 00:52:34,080
A SOFTWARE FOR THE DESIGN OF

1614
00:52:34,080 --> 00:52:36,720
NUCLEIC ACID STRUCTURES?

1615
00:52:36,720 --> 00:52:42,760
EXAMPLE, RATIONALLY DESIGNED

1616
00:52:42,760 --> 00:52:47,800
APTOMERS AND RIBOSOMES OR

1617
00:52:47,800 --> 00:52:48,040
DNAZYMES?

1618
00:52:48,040 --> 00:52:49,680
>>GIVEN A PARTICULAR
DOUBLE

1619
00:52:49,680 --> 00:52:50,920
STRANDED DNA SEQUENCE, MAKE A

1620
00:52:50,920 --> 00:52:51,440
BINDER TO IT.

1621
00:52:51,440 --> 00:52:52,640
WE'RE ACTUALLY MAKING A LOT OF

1622
00:52:52,640 --> 00:52:53,680
PROGRESS ON THAT PROBLEM.

1623
00:52:53,680 --> 00:52:54,920
SO THAT'S THE EASIEST PROBLEM

1624
00:52:54,920 --> 00:52:57,160
BECAUSE THERE, YOU KNOW WHAT

1625
00:52:57,160 --> 00:52:59,600
THE -- YOU HAVE A PRETTY GOOD

1626
00:52:59,600 --> 00:53:00,640
IDEA THAT'S B FORM DNA.

1627
00:53:00,640 --> 00:53:02,160
I THINK THE OTHER PROBLEMS ARE A

1628
00:53:02,160 --> 00:53:03,360
LITTLE BIT HARDER BECAUSE THE

1629
00:53:03,360 --> 00:53:05,880
STRUCTURE IS LESS REGULAR, BUT I

1630
00:53:05,880 --> 00:53:07,360
THINK THOSE ARE ALL CURRENT

1631
00:53:07,360 --> 00:53:09,440
PROBLEMS WHICH COULD BE

1632
00:53:09,440 --> 00:53:10,120
APPROACHED NOW.

1633
00:53:10,120 --> 00:53:11,320
AND I SHOULD SAY WE'RE ALWAYS

1634
00:53:11,320 --> 00:53:12,320
INTERESTED IN COLLABORATION, SO

1635
00:53:12,320 --> 00:53:14,320
IF PEOPLE HAVE THINGS WHERE

1636
00:53:14,320 --> 00:53:15,520
THEY'D LIKE BINDERS, LET ME

1637
00:53:15,520 --> 00:53:16,080
KNOW.

1638
00:53:16,080 --> 00:53:17,360
>>OKAY.

1639
00:53:17,360 --> 00:53:21,440
MAYBE ONE MORE AND IT'S A

1640
00:53:21,440 --> 00:53:22,640
SELF-DESCRIBED NAIVE QUESTION

1641
00:53:22,640 --> 00:53:23,800
BUT IT'S INTERESTING TO ME.

1642
00:53:23,800 --> 00:53:26,200
A NAIVE QUESTION, BUT CAN THESE

1643
00:53:26,200 --> 00:53:27,960
STRATEGIES BE USED TO PRODUCE

1644
00:53:27,960 --> 00:53:30,480
ARTIFICIAL BLOOD OR SPECIFIC

1645
00:53:30,480 --> 00:53:31,000
BLOOD TYPES?

1646
00:53:31,000 --> 00:53:33,040
>>WELL, I THINK WHAT THEY
CAN

1647
00:53:33,040 --> 00:53:34,400
BE -- THEY CAN BE USED FOR --

1648
00:53:34,400 --> 00:53:35,680
ONE OF THE USES OF THESE

1649
00:53:35,680 --> 00:53:36,880
ARTIFICIAL GROWTH FACTORS, IT'S

1650
00:53:36,880 --> 00:53:37,920
NOT QUITE IN ANSWER TO THAT

1651
00:53:37,920 --> 00:53:39,680
QUESTION, BUT YOU KNOW, SERUM IS

1652
00:53:39,680 --> 00:53:41,680
VERY -- IF YOU GROW CELLS FOR

1653
00:53:41,680 --> 00:53:43,560
ARTIFICIAL MEAT OR GROWING UP

1654
00:53:43,560 --> 00:53:45,000
T-CELLS FOR ADOPTIVE CELL

1655
00:53:45,000 --> 00:53:46,200
THERAPIES, YOU NEED TO ADD

1656
00:53:46,200 --> 00:53:47,560
GROWTH FACTORS AND CYTOKINES.

1657
00:53:47,560 --> 00:53:49,120
SO ONE OF THE THINGS WE CAN DO

1658
00:53:49,120 --> 00:53:50,080
IS SUBSTITUTE THOSE WHICH CAN BE

1659
00:53:50,080 --> 00:53:51,920
HARD TO MAKE AND EXPENSIVE AND

1660
00:53:51,920 --> 00:53:54,520
NOT VERY STABLE FOR MUCH CHEAPER

1661
00:53:54,520 --> 00:53:55,280
SYNTHETIC VERSIONS.

1662
00:53:55,280 --> 00:53:56,600
AS FAR AS BLOOD GOES, YOU KNOW,

1663
00:53:56,600 --> 00:53:59,200
YOU'D HAVE TO LOOK AT WHAT THE

1664
00:53:59,200 --> 00:54:01,600
ACTIVITIES -- WHAT ARE THE, I

1665
00:54:01,600 --> 00:54:03,040
GUESS HEMOGLOBIN, YOU WOULD NEED

1666
00:54:03,040 --> 00:54:04,240
SOMETHING THAT WOULD BE AN

1667
00:54:04,240 --> 00:54:04,840
OXYGEN CARRIER.

1668
00:54:04,840 --> 00:54:06,360
I THINK ONE COULD GO THROUGH IT.

1669
00:54:06,360 --> 00:54:07,560
I'M NOT ENOUGH OF AN EXPERT TO

1670
00:54:07,560 --> 00:54:10,680
KNOW WHAT THE KEY COMPONENTS ARE

1671
00:54:10,680 --> 00:54:11,880
THAT ONE WOULD NEED, BUT I THINK

1672
00:54:11,880 --> 00:54:12,880
IT'S A VERY INTERESTING

1673
00:54:12,880 --> 00:54:13,120
QUESTION.

1674
00:54:13,120 --> 00:54:14,200
>>OKAY.

1675
00:54:14,200 --> 00:54:15,120
WELL, IF THERE'S NOTHING ELSE

1676
00:54:15,120 --> 00:54:16,440
FROM HERE, I WANTED TO REMIND

1677
00:54:16,440 --> 00:54:18,520
PEOPLE THERE'S A RECEPTION JUST

1678
00:54:18,520 --> 00:54:19,160
OUTSIDE HERE.

1679
00:54:19,160 --> 00:54:22,640
NOT THE ONE ON FAS TEARS, THAT'S

1680
00:54:22,640 --> 00:54:24,040
A HOLIDAY PARTY AND YOU WOULD BE

1681
00:54:24,040 --> 00:54:25,920
CRASHING THAT, BUT JUST OUTSIDE

1682
00:54:25,920 --> 00:54:27,800
THESE DOORS.

1683
00:54:27,800 --> 00:54:29,000
SO THANK YOU FOR THIS GREAT

1684
00:54:29,000 --> 00:54:29,200
TALK.

1685
00:54:29,200 --> 00:54:29,600
>>OKAY.

1686
00:54:29,600 --> 00:54:30,040
THANK YOU.

1687
00:54:30,040 --> 00:54:38,240
[APPLAUSE]