>> GOOD AFTERNOON. I WANTED TO TO DO A BRIEF NOW FLAW REA INTRODUCTION FOR KEVIN WHITE FROM THE UNIVERSITY OF CHICAGO. I'VE KNOWN KEVIN FOR ABOUT A DECADE AND HAVE BEEN WATCHING WITH GREAT INTEREST THE INSTITUTE HE HAS ASSEMBLED IN SYSTEM BIOLOGY AT THE UNIVERSITY OF CHICAGO. HE COMES FROM A DEVELOPMENTAL BIOLOGY MOLECULAR GENETICS BACKGROUND, HAVING DONE HIS GRADUATE WORK WITH DAVE AT STANFORD. AND HE PASSES A KEY STEP. I HAD TO LOOK IN THE DEPTHS OF HIS CV TO SATE THAT WAS IDENTIFIED BY -- TO SEE IT THAT WAS IDENTIFIED BY A MUCH SENIOR THAN ME COLLEAGUE YEARS AGO. IF YOU SEE A PERSON THAT GOES TO DAVE'S LAB, WONDERFUL SCIENCE ALWAYS DONE THERE, WHO CAN GET TWO OR MORE PAPERS OUT OF DAVE, YOU'RE SEEING SOMEBODY WHO IS GOING TO SUCCEED BRILLIANTLY. THE SCIENCE WITH A ALWAYS GREAT BUT HE DIDN'T ENJOY PUTTING PAPERS TOGETHER. SO KEVIN DID THAT AND THEN WENT ON TO DO MANY OTHER THINGS FOLLOWING THE INTO HIS FACULTY POSITION AT YALE AND THEN ON TO CHICAGO. THE THREADS THAT BRING HIM HERE TO CANCER IS OFTEN WHAT MOLECULAR MODEL DOES, IT PUTS YOU INTO A PROBLEM YOU HADN'T PLANNED ON. HE WILL BE TALKING ABOUT HOW HE CAME TO CANCER. BUT THE OTHER THREADS THAT BRING HIM HERE ARE THOSE IN GENETICS, GENOMICS AND HIGH THROUGHPUT KINDS OF TECHNIQUES WHEN HE'S ALWAYS HAD AN AFFINITY FOR. SO HAVING MENTIONED THOSE SEVERAL THREADS IN HIS WORK NOW YOU CAN SEE THE REAL DEAL. KEVIN. >> THANKS VERY MUCH BARBARA FOR INVITING ME. I'VE APPRECIATED THE MEETINGS I'VE HAD SO FAR WITH PEOPLE AND I LOOK FORWARD TO CHATTING WITH MORE OF YOU OVER THE NEXT DAY AND-A-HALF. SO I THOUGHT I WOULD ACTUALLY START OUT BY SAYING A FEW MORE WORDS ABOUT THIS INSTITUTE THAT I FOUNDED IN 2006. SO THE IDEA BEHIND THIS, AND THE REASON I WANTED TO SAY A FEW WORDS IS BECAUSE I THINK ESPECIALLY IN SOME OF THE DISCUSSIONS THIS MORNING, IT SEEMS LIKE THE DIRECTIONS THAT WE'RE TAKING IN THE INSTITUTE FOR GENOMICS AND SYSTEMS BIOLOGY IN CHICAGO, VERY MUCH DOVE TAILED WITH THE SORTS OF THINGS THAT PEOPLE HERE ARE THINKING A LOT ABOUT THESE DAYS. AND I THINK THERE MIGHT BE SOME TREMENDOUS OPPORTUNITIES FOR COLLABORATION WITH PEOPLE IN THE INTRAMURAL PROGRAM HERE WITH US INTRAMURALITES. SO THE IDEA BEHIND THE INSTITUTE IS BASICALLY TO HARNESS THE DISCOVERY POWER OF GENOMICS AND SYSTEMS BIOLOGY APPROACHES, AND AS RAPIDLY AS POSSIBLE, PUSH BASIC DISCOVERIES INTO TRANSLATIONAL AND CLINICAL RESEARCH WITH THE IDEA OF ACCELERATING DIAGNOSTICS AND THERAPEUTIC DEVELOPMENT. YOU'VE PROBABLY SEEN LOTS OF SYSTEMS BIOLOGY TALKS THAT HAS DIFFERENT VERSIONS OF THIS KIND OF, YOU KNOW, OVERLAPPING DIAGRAM OF BIOLOGY, COMPUTATION AND TECHNOLOGY DEVELOPMENT. REALLY IN A REAL WAY, THESE OVERLAPPING CIRCLES ARE BECOMING MORE AND MORE INTERMESHED. WE'RE ALL BEGINNING TO REALIZE THAT. AND AS I SAID, OUR GOALS ARE TO DEVELOP ON THE CLINICAL SIDE MORE DIAGNOSTIC THERAPEUTIC APPROACHES THAT ARE DATA DRIVEN FROM THE STANDPOINT OF BASIC BIOLOGICAL WORK. AND WE ALSO HAVE A BRANCH OUT AT ARGON NATIONAL LABORATORY THAT'S CONNECTED WITH THE DEPARTMENT OF ENERGY THAT'S VERY FOCUSED ON ENVIRONMENT. AND I'M NOT PERSONALLY DOING RESEARCH IN THAT AREA, BUT WE HAVE THREE OR FOUR INVESTIGATORS OUT THERE WHO ARE VERY ACTIVE. I WON'T BE TALKING ABOUT THAT TODAY. WE HAVE SOUTHERN CORE FACULTY, INCLUDING MYSELF. FIVE AT THE UNIVERSITY OF WHICH, TWO AT ARGON. THERE ARE A LOT OF AFFILIATES OF OUR OPERATION. WE'VE MADE A GREAT AMOUNT OF EFFORT TOWARD TRYING TO INTEGRATE OUR RESEARCH PROGRAMS WITH THE CLINICAL AND TRANSLATIONAL RESEARCHERS, BOTH AT THE UNIVERSITY OF WHICH AND BEYOND. AND SOME OF THOSE FACULTY ARE OFFICIALLY FELLOWS AND PARTICIPATE IN A LOT OF OUR ACTIVITIES. WE HAVE A BUDGET THAT WE'VE, PROBABLY ABOUT 10% OF THIS IS, WELL NOT 10% OF THIS BUT ANOTHER 10% OF THAT NUMBERS COMING FROM INTERNAL FUNDING. AND REALLY, THE HEART OF WHAT WE DO IS CENTERED AROUND THE GRADUATE STUDENTS, THE POST DOCS. WE DO HAVE EDUCATIONAL OUTREACH PROGRAMS THAT ARE PART SEVERAL CENTERS THAT WE HAVE, INCLUDING NIH NATIONAL CENTER FOR SYSTEMS BIOLOGY, TH CONTE CENTER FOCUSED ON NEUROPSYCHIATRIC DISORDERS AND CENTER FOR DATA MINING WHICH GOT CUT OFF BY OUR LOGO THERE. OF PARTICULAR INTEREST FOR TODAY AND THE NATIONAL CANCER INSTITUTE, IF YOU LOOK AT OUR FACULTY MEMBERSHIP, IT'S ABOUT HALF OF THE FACULTY ARE OVERLAPPING WITH THE UNIVERSITY OF WHICH COMPREHENSIVE CANCER CENTER. AND THIS IS JUST BY WAY OF SAYING THAT CANCER HAS BEEN A MAJOR FOCUS FOR US, ALTHOUGH AS I JUST MENTIONED THERE ARE OTHER AREAS, AT LEAST 50% OF THIS INSTITUTE'S EFFORT IS FOCUSED ON CANCER. AND ONE OF THE MOST SUCCESSFUL PROGRAMS THAT WE'VE LAUNCHED IS ACTUALLY TRAINING OF CLINICAL RESEARCH FELLOWS, PEOPLE WHO HAVE COMPLETED THEIR M DVMENTZ THEIR MD PH.D.'S AND THEY COME AND SPEND TIME IN THE LAP TREELABORATORIES AT THE INSTITUTE. THIS IS JUST A SUBSET OF THE CLINICAL RESEARCH FELLOWS. AND THIS SUBSET OF FELLOWS ARE COORDINATING A PROJECT OR SERIES OF PROJECTS, REALLY, THAT COME UNDER THE HEADING OF THE CHICAGO CANCER GENOMES PROJECT. AND THE IDEA BEYOND THE CHICAGO CANCER GENOMES PROJECT, AND THEN I'LL GET ON TO THE SCIENCE, IS THAT LEVERAGING THE GREAT EFFORTS THAT ARE BEING PUT INTO THINGS LIKE THE CANCER GENOME AT ATLAS AND THE TARGET CONSORTIA AND SO FORTH. WHAT WE DO IS WE WORK VERY CLOSELY WITH SETS OF INVESTIGATORS WHO ARE DOING CLINICAL STUDIES, SOME IN CLINICAL TRIAL STUDIES, SOME OF THEM ARE, JUST HAVE INTERESTING SETS OF PATIENTS WHERE THEY'RE TRYING TO IDENTIFY GENETIC DIFFERENCES THAT MIGHT CORRESPOND TO THERAPIS THERAPY OR SUCH. AND WE COLLECT GENOME DATA. WE PRODUCE THIS IN HOUSE AND THESE CLINICAL FELLOWS ARE SORT OF ACTING AS HOBBS FOR THIS -- HUBS FOR TH IS NETWORK OF CLINICAL AND TRAPS LATIONAL INVESTIGATIONS THAT WE'VE BEGUN. AND THEY ARE BACKED UP BY A SERIES OF CORE FACILITIES THAT DO EVERYTHING FROM THE SEQUENCING TO THE BIOINFOMATICS, PLATFORMS THAT A LOT OF THE THINGS THAT WE DO EXIST ON. TO SOME MORE NOVEL THINGS. WE HAVE SOME DNA CORE THAT WILL MAKE UNIQUE SETS OF REAGENTS FOR PEOPLE. WE HAVE A MICRO WESTERN CORE WHICH IS BASED ON A TECHNOLOGY WHICH WAS DEVELOPED BY ONE OF OUR JUNIOR FACULTY. AND WE HAVE DRUG SCREENING AND MAC COSO FORTH AS WELL. I GUESS THE HIGH LEVEL CONCEPT, AGAIN MOST OF YOU HAVE PROBABLY SEEN SLIDES LIKE THIS, IS THAT WE'RE TRAC TRIANGLATING TO BUILD PREDICTED MODELS FOR COMPLEX BIOLOGICAL PHENOMENA. WHAT I'M GOING TO BE TALKING ABOUT TODAY IS HOW I'LL START OUT WITH AN EXAMPLE OF BUILDING A NETWORK TO DESCRIBE PART OF THE -- DEVELOPMENTAL PROGRAM AS A DEMONSTRATION PROJECT AND HOW THAT LET US TO A NEW KIDNEY CANCER GENE. AND IF I MAKE IT ALL THE WAY THROUGH THE TALK, I'M GOING TO SHOW YOU AN EXAMPLE HOW WE'VE GONE ALL THE OTHER WAY AROUND AND STARTED WITH JEESM SEQUENCE DATA AND ENDED UP FRUIT FLIES TO VALID A NEW SUPPRESSOR GENE. S MOSGENE. MOST PEOPLE HAVE ALL SEEN THIS ORGANISM. WE HAVE A LOT IN COMMON WITH FRUIT FLIES. ABOUT A DECADE AGO WHEN THE FRUIT FLY GENOME WAS SEQUENCED, IT BECAME VERY QUICKLY APPARENT THAT CLOSE TO 70% OF THE GENES IN THE FRUIT FLY GENOME HAVE SOME RECOGNIZABLE COUNTERPART IN IT. AND OF COURSE IN CANCER BIOLOGY, WE ALL NOW, ALL OUR STUDENTS WHO COME INTO OUR LABS NOW TAKE FOR GRANTED THAT THE MOLECULAR PATHWAYS IN THIS FRUIT FLY ORGANISM ARE HIGHLY CONSIDERED WITH WHAT'S FOUND IN US. SO AS A STARTING MONEY, I'M GOING TO MENTION ONE OF THE PROJECTS GOING ON IN OUR INSTITUTE. THIS IS ONE THAT I ACTUALLY HEAD UP WHERE SYSTEMATICALLY MAPPING THE REGULATORY ELEMENTS IN THE FRUIT FLY GENOME. AND SO THE REASON I'M STARTING HERE, BESIDES JUST TO SAY YOU CAN GO TO SOME WEBSITES AND GET LOTS OF GREAT DATA AS A STARTING POINT FOR IDENTIFYING WHERE THE REGULATORY ELEMENTS ARE IN THIS ORGANISM'S GENOME, IS THAT WE CAN ACTUALLY USE THIS KIND OF DATA ALSO AS A STARTING POINT TO INVESTIGATE COMPLEX BIOLOGICAL PROBLEMS. ONE OF THE MOST HEAVILY STUDIED AREAS OF -- GENETICS IS HOW THE EARLY EMBRYO SETS UP ITS PATTERNING. OVER THE LAST SEVERAL YEARS WE'VE COME TO UNDERSTAND A GREAT DEAL ABOUT THE PROCESS. THE REGULATORY NETWORK STARTS WITH THE DEPOSITION OF MATERNAL GENES FROM THE FORMED GRADIENTS FROM THE ANTERIOR TO POSTERIOR. AND THESE FACTORS IN TURN REGULATE A SET OF GENES KNOWN AS THE GAS GENES. THE GAS GENES ARE BASICALLY ALL TRANSCRIPTION FACTORS. THEY ARE ALL TRANSCRIPTION FACTORS. THEY ARE SUCH NAMED BECAUSE WHEN YOU MUTATE ONE OF THESE TRANSCRIPTION FACTORS YOU GET GAPS IN MULTIPLE SEGMENTS IN THE EMBRYOS. THESE ARE KNOWN TO REGULATE A SET OF -- GENES. THEY ARE AGAIN TRANSCRIPTION FACTORS. THEY SET UP A FINER DELIBERATATION OF THE BODY PLAN. AND THE TARGETS OF THESE PAIR-RULE GENES ARE RESPONSIBLE FOR SETTING UP THE POLARITY UPON THE EMBRYO. THESE TURN OUT TO ENCODE FACTORS THAT ARE IMPORTANT IN MAJOR PATHWAYS THAT ARE CONSERVED IN HUMAN, ARE KEY REGULATORY PATHWAYS IN A WIDE VARIETY OF CANCERS, INCLUDING HEAD SIGNALING, PGF BETA SIGNALING AND SO FORTH. AND SO THE SORT OF LOGIC GOES THAT IF WE MAP THE FAWRG TARGETS OF THESE PAIR-RULE GENES AND OTHER GENES IN THIS NETWORK, THEN WE MIGHT FIND ADDITIONAL GENES THAT ARE IMPORTANT IN CANCERS. THAT'S A PRETTY BIG STEP. AND THE FIRST STEP WE WANT TO TAKE WAS COULD WE JUST FIND MORE GENES BECAUSE THEY'RE INVOLVED IN THIS PROCESS. AND SO WHAT WE BEGAN OUR FOCUS ON, AND THIS STARTED BEFORE THE -- PROJECT STARTED SO THIS PROJECT STARTED SEVEN YEARS AGO OR EIGHT YEARS AGO. WE STARTED WITH -- THESE ARE HOMEOVOHOMEHOMEOVOX. SO WHAT ONE CAN DO, THEY CAN TAKE THE MUD ENCODE DATA MAPPING FLUX AND YOU CAN OVERLAY THAT ON DATA LIKE THIS WHERE WE'VE LOOKED AT DIFFERENTIAL GENE EXPRESSION IN EITHER OR -- AND YOU CAN FILTER THE GENES FROM THOSE THAT HAVE BINDING SITES NEARBY AND THOSE BECOME FOR EXAMPLE YOUR PUNITIVE DIRECT TARGETS OF THESE TWO FACTORS. SO THAT SORT OF LOGIC IS FAIRLY STRAIGHTFORWARD. THEN YOU HAVE TO GO ABOUT PROVING THAT EACH AND EVERY ONE OF THOSE ARE TARGETS, IF THOSE WHAT YOU SO WANT TO DO. BUT INSTEAD WHAT WE TEND TO DO IS ADD MORE DATA TO TRIANGULATE MORE, TO TRY TO FIGURE OUT WHICH OF THESE TARGETS MIGHT BE BIOLOGICALLY IMPORTANT. ONE CAN ADD PROTEIN PROTEIN INTERACTION DATA. WE'VE MADE QUITE A BIT OF USE OF LITERATURE MINING DATA. WE WORK WITH ANDRE -- WHO BUILDS NETWORK DATA BASES THAT ARE BASED ON NATURAL LANGUAGE PROCESSING OF LITERATURE AND THEN HE GETS A FIRST APPROXIMATION OF A NETWORK AND WE GO IN AND VALID IT. AND WITH ANDRE'S GROUP, WE SORT OF NOW A NUMBER OF YEARS AGO CAME UP WITH AT LEAST ONE WAY. NOW PEOPLE HAVE TAKEN A NUMBER OF WAYS TO THIS PROBLEM OF INTEGRATING HETEROGENEOUS DATA SETS. EACH ONE OF THESE DATA SETS IS INHERENTLY NOISY, IF YOU DO AN RNAC EXPERIMENT OR A CHIP EXPERIMENT. THERE'S A FAIR AMOUNT OF NOISE SO YOU HAVE TO FILTER. SO YOU CAN FILTER THE NOISE HERE BEST IN EACH ONE OF THOSE TYPES OF EXPERIMENTS. AND YOU CAN COME UP WITH SOME KIND OF CONFIDENCE SCORE. SO WHAT THE METHOD THAT WE USE SORT OF MAKES USE OF IS SORT OF COMBINING CONFIDENCE SCORES ACROSS, NOT SORT OF BUT ACTUALLY, COMBINING CONFIDENCE SCORES ACROSS THESE HETEROGENEOUS DATA SETS. SO YOU CAN BUILD JOINT PROBABILITIES BASED ON THIS. THEN WHAT ONE CAN DO IS USE SIMULATIONS TO ESSENTIALLY LAY DOWN A NETWORK. SO YOU START OUT WITH ONE NETWORK. MAYBE IT HAS A THOUSAND FACTORS IN IT AND 2000 INTERACTIONS OR 4,000 INTERACTIONS OR WHATEVER NUMBER OF INTERACTIONS YOU WANT TO START WITH. AND THEN YOU CAN REMOVE A NODE, YOU CAN CHANGE CONFIGURATION OF AN EDGE, YOU CAN ADD AN EDGE WHERE THE GENES AND THE PROTEINS ARE NODES AND INTERACTIONS ARE EDGES, SO PROTEIN DNA INTERACTION MIGHT BE ONE EDGE. EXPRESSION CORRELATION MAY BE ANOTHER. PROTEIN PROTEIN MAY BE ANOTHER. AND YOU JUST KEEP RECALCULATING THIS JOINT PROBABILITY HOVER AND -- HOVER AND OVER AGAIN FOR VARIOUS CONFIGURATIONS OF THIS NETWORK AND YOU TYPICALLY DO THIS ON HIGH PERFORMANCE COMPUTING, EXPLORE LARGE AMOUNT OF SPACE. AND YOU COME UP WITH A BIG HAIRBALL KIND OF FIGURE LIKE THIS. WHAT'S UNDERNEATH EACH ONE OF THESE EDGES IS A JOINT PROBABILITY SCORE DERIVED FROM BUILDING THIS NETWORK IN MILLIONS AND MILLIONS OF DIFFERENT WAYS. AND SO YOU CAN, IN THIS SYSTEM, IF IT WERE LOGIC, YOU WOULD CLINICLICK ON IT AND SEE THE EVIDENCE. WE HAVE PF BINDING, LITERATURE IN HYBRID DATA BUT NOW WE HAVE MORE OTHER KIND OF PROTEIN PROTEIN INTERACTION DATA. I WANT TO DRAW YOUR ATTENTION TO IS THIS ONE FACTOR WHEN FROM ITS ENTIRE NETWORK. THAT'S WHAT I'M GOING TO TALK ABOUT. FOR THE FIRST STORY THAT I'M GOING TO TELL YOU. SO HERE, THIS FACTOR WHEN WE FOUND IT WAS CALLED CG9924. IT INTERACTS WITH PUCKERED. IT'S A [INDISCERNIBLE] PHOSPHATASE THAT IS UPSTREAM OF GENE KINASE. THIS IS SHOWN IN THIS PART OF THE NETWORK THAT'S NOT VISIBLE TO ME BUT HOPEFULLY VISIBLE TO YOU. CG9924 IS A DIRECT TARGET OF -- AND A LOT OF OTHER FACTOR IN THE NETWORK. IT INTERACTS WITH SO MANY FACTORS IN THE NETWORK IF YOU WERE TO GO FROM ANY ONE POINT IN THE NETWORK TO ANY OTHER POINT IN THE NETWORK, YOU MOST OFTEN HAVE TO PASS THROUGH THIS CG9924 IN ORDER TO GET THERE. OKAY. SO THAT'S WHAT THIS CONNECTIVITY SCORE IS. IT'S THE MOST INTERCONNECTIVE COMPONENT OF THIS NETWORK SO THAT WE ASK FOR ALL TWO COMPONENTS, ANY TWO COMPONENTS OF THE NETWORK, WHAT'S THE FACTOR YOU HAVE TO PATS THROUGH THE MOST AND THEN STOP THEM. OTHER THINGS TOWARD THE TOP ARE HOMEODOMAIN PROTEINS AND A LOT OF IMPORTANT BIOLOGICAL STUFF. IN HUMAN, THIS ALREADY HAD A NAME, IT'S CALLED SPOP AND THE NAME DOESN'T HAVE ANYTHING TO DO WITH THE FUNCTION. SO I'M NOT GOING TO TALK ABOUT WHAT THE NAME MEANS WE JUST CALL IT SPOP. IT'S BETWEEN THE -- HUMAN AT A VERY HIGH LEVEL WHICH ALSO INTERESTS US. IT'S EXSUPRESSED IN STRIPES AND IN THOUGHTS YOU LOSE THE STRIPES. IT CONTAINS BOTH A MASS AND B TO B DOMAIN. THIS TURNS OUT TO BE PREDICTIVE TO BE PART OF A CUT 3 LIE GASE COMPLEX. I'LL GET TO THAT IN A MOMENT. IT LOCALIZES TO THE SEGMENTAL GROOVE. THIS IS WHERE THE EMBRYO IS ACTUALLY PHYSICALLY MAKING THE SEGMENT, SO A LOT IS KNOWN ABOUT THE DEVELOPMENTAL BIOLOGY OF THE TRANSCRIPTION REGULATORY NETWORK BUT NOT SO MUCH ABOUT THE MORPHOLOGICAL CONTROL AT A PHYSICAL LEVEL. AND SO SOMEHOW THIS IS LOCALIZED TO THE GROOVES AND I'M NOT GOING TO TALK TOO MUCH MORE ABOUT THAT OTHER THAN TO POINTE POINT OUT THAT ANOTHER GROUP HAD IDENTIFIED THIS AND WE AT THE SAME TIME HAD DONE A NUMBER OF MUTANT STUDIES WHERE WE GET THE SAME KIND OF RESULTS. THEY HAD NAMED IT ROAD KILL AND IT'S BECAUSE OF THE PHENOTYPE LOOKS LIKE THIS. SO THIS IS YOUR NORMAL CUTICLE PREP AND THIS IS A ROAD KILL SPOP MUTANT. YOU CAN SEE, YOU GOT THESE FURRY LITTLE IDENTICAL GROWTHS HERE. THAT'S ASSOCIATED WITH SEGMENTAL PHENOTYPES. THIS WASN'T ORIGINALLY IDENTIFIED IN THE WILLHART AND WISHOFF SCREEN THAT DECIPHERED MOST OF THE MEMBERS, KNOWN MEMBERS OF THE THIS PATHWAY IN PART BECAUSE THE PHENOTYPE IS SO EXTREME. SO IT'S NOT A VERY GOOD PHENOTYPE TO DO ADDITIONAL GENETICS WITH. AND SO WHAT WE DID IS WE TURNED TO THE -- I. SO WE PREDICTED IT INTERACTED WITH PUCKERED. IT'S PART OF THE GENE KINASE SIGNALING PATHWAY. SO WE TAKE TNF AND WE CAN OVEREXPRESS TNF IN THE EYE AND OVERPRODUCE APOPTOSIS. IF YOU WANT TO SEE A GENETIC INTERACT IN THIS PATHWAY, YOU CAN RULES ITS COPY NUMBER BY ONE. SO IT'S HETERO ZYGOTIC, AND IN THIS CASE, WE GOT A WEAK EXPRESSING ISO IF THE EYE GOES AWAY WHAT'S HAPPENING IS ACTIVATING PUCKERED IN THIS CASE. IF THE I COMES BACK IT'S INHIBITING PUCKERED BECAUSE OF THE LOGIC OF THE GENETICS, OKAY. AND SO WHAT HAPPENS IS THE I COMES BACK IN THIS HIGH EXPRESSING TNF FLY I. SO THAT'S ADJUSTING IT THE INHIBITOR OF PUCKERED WHICH IN TURNS INHIBITS GENE KINASE SO IT'S AN ACTIVATOR OF THE PATHWAY NORMALLY. SO THIS IS A LOTS OF FUNCTION. THAT'S WHAT WE THINK IT'S DOING GENETICALLY. IT'S PART OF THE A -- LIGASE COMPLEX. THIS IS JUST A STUDY, ONE GEL SHOWING THAT NORMALLY YOU HAVE PUCKERED IN TWO FORMS. IF YOU ADD SPOP INTO FLY CELLS YOU GET A REDUCTION AND IF YOU ADD AN INHIBITOR OF THE PROTOJOE SO MANPROTOJOE -- SO SOME. I DON'T THINK I'M GOING TO SHOW YOU THE GELS BUT WE WENT ON TO SHOW THAT THIS IS A CONSERVED FUNCTION IN HUMAN AND THAT IT'S REGULATING THE KINASE SIGNALING MATT WAY. SO WHAT DOES THAT HAVE TO DO WITH CANCER. WE MADE A LEAP OF FAITH AND WE DECIDE TO DO ASK WHETHER THIS SPOP PROTEIN THE HUMAN VERSION WAS ASSOCIATED WITH ANY CANCER TYPES. SO WE USED CANCER MICRO RAYS WITH ABOUT 20 DIFFERENT TUMOR TYPES ON THEM AND WE ASKED WHETHER ITS LEVELS OF EXPRESSION WERE ASSOCIATED WITH NORMAL TISSUE. WHAT WE FOUND IS THAT KIDNEY CANCERS IN PARTICULAR AFTER WE MADE A NEW ARRAY WITH A NUMBER OF DIFFERENT TUMOR TYPES ON IT, WE FOUND THAT CLEAR CARCINO CARCINOMAS ARE ALMOST ENTIRELY EXPRESSING HIGH LEVELS OF SPOP WHERE THE NORMAL KIDNEY TISSUE IS NOT. FURTHERMORE, WE FIND THAT IN CLEAR CELL RCCs THAT SPOP IS MISLOCALLIZED TO THE NUCLEUS AND THAT'S SHOWN, YOU CAN SEE THAT A LITTLE BIT HERE IN THIS HISTO PATHOLOGY SLIDE. BUT IN B, THESE ARE KHAKI CELLS WHICH ARE CLEAR CELL CARCINOMA CELL LINE. THE SPOP IN THE CYTOPLASM AS WELL AS SOME NUCLEAR STANDING. IN HECK 293 CELLS YOU DON'T SEE THIS. SO IF ONE THEN ASKS A QUESTION WHAT HAPPENS IF WE TAKE SPOP AND WE CLIP ITS LOCAL NUCLEAR DOUGH MODEL DOMAIN. IT HAPPENS A LOCAL DOMAIN AT ITS C TERMINAL END AND PUT IT BACK INTO CELLS. IF WE PUT FULL LENGTH SPOPS INTO CELLS ITEMS TRIGGERING -- AND THESE DO GO TO THE CYTOPLASM, YOU GET EXCESS PROLIFERATION. IF YOU KNOCK IT DOWN IN HILO CELLS, YOU SEE VERY LITTLE OR NO DIFFERENCES IN CELL DEATH. IF YOU KNOCK IT DOWN IN THE KHAKI TWO CLEAR CELL AWE REA NOW CELL CARCINOMAS YOU SEE AN INCREASE IN CELL DEATH. RECENTLY CELL LINES NECESSARY AND SUFFICIENT FOR SURVIVAL OF THESE CANCER CELLS AND IT'S EFFICIENT TO PUSH HEX CELLS OR HILO CELLS TO BE EVEN MORE PROLIFERATIVE. SO WE THEN WENT TO TAKE THE PSYCHO PLASMIC SMOP AND ASK WHETHER WE ARE PROMOTE FEUMPLE GENERAL JUST IN THE XENO GRAPH MODEL WITH SUBCUTANEOUS INJECTION AS LONG WITH HOMO AUTOPSIY IMPLANTATIONS. WHAT THIS IS SHOWING YOU HERE IS IF YOU TAKE CONTROLS OR IF YOU TAKE THE FULL LENGTH SMOP AND YOU PUT THESE INTO HUMAN EMBRYONIC KIDNEY CELLS YOU DON'T GROW TUMORS AFTER SIX WEEKS BUT IN THE SPOP NUCLEAR LOCALIZATION MINUS WHERE IT'S SIGH HE TOE PLASMIC YOU GROW THESE TUMORS. IF YOU PUT THEM IN SUBCAPSULARLY AGAIN YOU GET A VERY HIGH RATE OF TUMOR TAKE IN THESE. AND THEY'RE VERY HIGHLY VASCULARIZED, THESE TUMORS. SO WITH THE HELP OF BRENDA, WE ACTUALLY, THIS IS A FIRST IN CLASS INTERACTION WITH KRUL 3 AND SHE'S VERY INTERESTED IN THE BIOCHEMISTRIES AND CRYSTALLOGRAPHY OF THESE COMPLEXES. SO THEY WERE EAGER TO CO-CRYSTALLIZE OUR SPOP MOLECULE WITH THE SUBSTRATES. THIS IS JUST SHOWING THE CRYSTAL STRUCTURE FOR ONE OF THE SUBSTRATES PUCKERED ALONG WITH THE SPOP AND I'M POINTING TO THE CARTOON BECAUSE THIS IS MORE COMFORTABLE FOR ME. I JUST WANT TO POINT OUT IS THE INTERFACE WE IDENTIFIED AND CALLED THE SPOP BINDING CONSENSUS WHERE WE SEE MULTIPLE SUBSTRATES NOT JUST PUCKERED ARE BINDING AT THIS INTERFACE IN THE MAP DOMAIN. AND THIS PARTICULAR INTERFACE HAS THIS DEGENERAL RU DEGENERAL RUT DERATE DEGENERATE MOW TEST. THEY ARE BINDING TO THIS CITE. WE LOOK FOR THIS SITE AMONG THE GENES THROUGHOUT THE HUMAN GENOME. WE MADE OUR LIST AND WE'RE INTERESTED IN ASKING WHAT ARE THE POSSIBLE TARGETS OF SPOP IN THESE KIDNEY CANCER CELLS ULTIMATELY WHAT IS WE TO NOT KNOW. IT'S GOING TO CYTOPLASM. WHO IS THIS TARGETING. AMONG THE PHOSPHATASES THAT MIGHT BE INVOLVED IN IDENTIFYING USP6 AND 7. WE ALSO IDENTIFIED 310. AND TO MAKE A LONG STORY SHORT, P10 IS INDEED A TARGET OF SPOP. WE'VE TAKEN IT DOWN TO THE CO-CRYSTAL STRUCTURE YOU CAN SEE WHEN SPOP IS HIGH IN THE CLEAR CELL CARCINOMAS, P10 IS LOW. AND SO WE THINK THAT ONE, PART OF THE MECHANISM OF HOW SPOP MAY BE CAUSING HYPER PROLIFERATION IS BY DOWN REGULATING P10 DIRECTLY BY CAUSING IT TO BE DEGRADED. BUT WE ALSO SEE THIS WITH USP WITH GLIAL 2 WHICH IS PART OF THE -- PATHWAY AND WITH DAX WHICH IS A PROAPOPTOTIC MOLECULE. THIS IS SHOWING SOME ADDITIONAL BLOTS FOR DAX WHERE WE KNOCK DOWN AND SEE INCREASE OF DAX. WE SEE 184 PHOSPHORYLATION. SO THIS IS JUST BY WAY OF SAYING THAT WE THINK THE SPOP WHEN IT'S THE CYTOPLASM IN THE CONDITION IS AFFECTING THIPTIONZ LIKE USB6, 7DAX. THIS IS BLOCKING APOPTOSIS. P10 IS DOWN REGULATED. WE ALSO SEE THAT WHEN WE KNOCK DOWN GLEE ARTIFICIALLY WITH RNASE WE GET HYPER PROLIFERATION. WE THINK IT'S ACTING NOT ONLY IN FLY AS A HUB BUT ALSO KIDNEY CANCER AS A HUB. SPECIFICALLY WHEN IT GETS MISLOCALLIZED TO THE CYTOPLASM. SO THOSE OF YOU THAT KNOW SOMETHING ABOUT KIDNEY CANCER KNOW THAT STRVMENT HL MUTATIONS ARE THE MOST COMMON GENETIC ABNORMALITY IN KIDNEY CANCERS. AND SO WE NATURALLY ASK WHETHER VHL WHICH REGULATES HIF IS INVOLVED. AND SO WHAT WE SEE HERE IS WHEN WE MAP HIF GENOME WIDE FOR BINDING SITES, WE HAVE POTENTIAL REGULATORY REGIONS IN AND AROUND SPOP GENE. AND IF YOU EXPOSE THESE HAX CELLS TO HIGH POP TICKER CONDITIONS, IF YOU KNOCK DOWN HIF TO ALPHA IN PARTICULAR YOU GET KNOCK DOWN OF SPOP. AND IF YOU OVER EXPRESS HIF TO ALPHA HERE ON A FLY TAG, YOU SEE UP REGULATION OF S IN POP. WHAT WE THINK IS HAPPENING IN VHL MUTANTS, HIF IS GOING TO THE NUCLEUS. IT'S ACTIVATING SPOP. IT'S ALSO KNOWN TO ACTIVATE -- AND WE CONSIDER THIS A PRETTY NASTY LOOP BECAUSE YOU THEN CREATE OVER EXPRESSION OF SPOP EITHER PASSIVELY OR ACTIVELY ACCUMULATES IN THE CYTOPLASM. TRIGGERS HYPER PROLIFERATION. OVERGROWTH OF CELLS WHICH CREATES LOWER OXYGEN AMOUNTS AND GET MORE HIF, MORE VEG F AND SO ON. SO THAT'S SORT OF OUR MODEL. WE'RE IN THE PROCESS NOW WHERE WE'VE MADE SPOP INDUCIBLE KNOCK DOWNS AND WE'RE STARTING TO DO MOUSE MODELS, WE'RE TRYING TO SEE IF WE CAN INDUCE RNAI TO MOK DOWN SPOP -- TO KNOCK DOWN SPOP IN MOUSE MODELS OF KIDNEY CANCER AND SEE IF WE CAN HAVE AN EFFECT ON THE DISEASE I WISH I COULD TELL YOU THEY HAD THE RESULTS. BUT THE MICE ARE EATING THEIR CHOW RIGHT NOW. OKAY. SO IN THE NEXT STORY I WANT TO TALK ABOUT AND AGAIN I SHOULD SAY -- OOPS, WHAT HAPPENED? I GUESS I'M DONE. THERE I GO. OKAY. I WANT TO SAY THAT YOU KNOW, THIS WORK HAS ALL BEEN DONE AS A COLLABORATION BETWEEN BASIC SIGN SCIENTISTS IN OUR SCIENTISTS IN OUR LAB AND TRANSLATIONAL GROUPS PARTICULARLY NOW WE'RE MOVING INTO MOUSE MODELS WE'RE GETTING A LOT OF HELP FROM OUR COLLEAGUES IN NEUROLOGY TO MOVE THIS FORWARD. AND WE THINK THIS MAY BE SOMETHING TO THINK ABOUT IN TERMS OF DEVELOPING SMALL MOLECULES. WE KNOW THAT THE INTERFACE WE WOULD LIKE TO HIT WE'RE EXPLORING VARIOUS WAYS TO DO THAT. SO ONE COULD THINK ABOUT SOMETHING LIKE COMBINING SPOP INHIBITORS WITH VEGF INHIBITORS AS SOMETHING WORTH FOLLOWING. SO I WANT TO NOW TURN TO A DIFFERENT CANCER TYPE. AND A DIFFERENT KIND OF A PROBLEM. AND THIS IS ONE THAT PEOPLE HERE ARE VERY WELL AWARE OF. A LOT OF THIS TYPE OF WORK HAS HAPPENED RIGHT HERE IF IN PEOPLE'S LABS OVER THE LAST DECADE OR SO. AND THIS IS A PROBLEM OF TAKING HETEROGENEOUS DISEASE AND TRYING TO DECIPHER WHAT THE SUBTYPES ARE BY USING MOLECULAR MARKERS. I WANT TO TELL YOU A COUPLE WAYS THAT WE'RE APPROACHING THIS ACTUALLY IN BREAST CANCER. ONE OF THE WAYS IS TO FOCUS ON ACTUALLY THE NUCLEAR RECEPTORS. AND TRY TO IDENTIFY SUBSETS OF PATIENTS THAT MIGHT RESPOND TO CERTAIN THARYPTION AGAINST -- THERAPIES AGAINST NUCLEAR RECEPTORS BETTER THAN OTHER SUBSETS OF PATIENTS. AND WE'RE GOING FOR THE HOLY GRAIL AND TRYING TO UNDERSTAND THE UNDERLYING GENETICS AND HOW THE UNDERLYING GENETICS OF THESE TUMORS IS RELATED TO THE CELLULAR STATE OR ABERRANT CELLULAR STATES THAT ONE SEES. IT'S OUR FOCUS ON THIS PROBLEM OF ER POSITIVE VERSUS TRIPLE NEGATIVE BREAST CANCERS. AND AS YOU ALL KNOW, ER POSITIVE BREAST CANCERS HAVE MUCH BETTER OUTCOME IN PART BECAUSE THE DISEASE IS LESS AGGRESSIVE AND IN PART BECAUSE WE HAVE A LOT MORE TOOLS TO GO AFTER IT. TRIPLE NEGATIVE IS DEFINED BY WHAT IT DOESN'T HAVE. SO WE AND OTHER PEOPLE LIKE MILES BROWN SORT OF REASONED A WHILE AGO THAT IT WOULD MAKE SENSE TO START LOOKING AT THE HE IESTROGEN RECEPTOR TARGETS GENOME WIDE TO GET A BETTER UNDERSTANDING WHAT THE MECHANISMS OF ACTION ARE DURING PROLIFERATION OF ESTROGEN-DEPENDENT CANCERS. BUT WHAT I WANT TO TALK TO YOU ABOUT IS HOW WE'RE SORT OF USING THIS DATA, THIS KIND OF DATA IN ORDER TO TRY TO STRATIFY DISEASE. AT ADDITIONAL LEVELS. SO AS YOU KNOW, ESTROGEN RECEPTOR'S IMPORTANT -- IS ALSO IMPORTANT. PROLIFERATION OF -- CELLS IN BREAST CANCER IS DRIVEN BY BOTH OF THESE MOLECULES AND ESTROGEN RECEPTOR DIRECTLY IS THOUGHT TO REGULATE CMYC. TAKING THE SAME KIND OF INTEGRATIVE APPROACH, I'M NOT GOING TO WALK YOU THROUGH IT BUT TAKING THE SAME KIND OF APPROACH AS DROSOPHILA, WE ARE LOOKING AT VARIOUS TYPES OF DATA INCLUDING CHIP CHIP AND CHIP SEAT DATA EXPRESSION DATA. HERE WE'VE DONE A WHOLE GENOME RNAI SCREEN WHERE WE'VE LOOKED AT THE GENES THAT ARE NECESSARY FOR ESTROGEN DEPENDENT GROWTH AND FOR NON-ESTROGEN DEPENDENT GROWTH IN MCL7 CELLS. THIS IS JUST A SUBSET OF THOSE SCREEN AND YOU CAN'T SEE THE LABELS BUT THERE ARE A BUNCH OF CHROMCHROMATIN-RELATED PROTEINS THAT HAVE BINDING SITES THAT ARE REGULATED AND IN SOME CASES ARE IN EFFECT. WE MAPPED THESE INTO NETWORKS, AND WITH AT LEAST ONE OF THESE FACTORS, H2AZ, WE SHOWED A WHILE AGO THAT WE CANNOT ONLY VALIDATE IT'S A DIRECT TARGET IN THIS CASE OF MYC. WE FOUND TWO E-BOXES OF STREAM OF H2AZ. WITHIN 45 MINUTES YOU START TO GET AN ACCUMULATION OF MCY. IT TAKES SIGNIFICANTLY LONGER TO START TO ACCUMULATE H2AZ IN THESE CELLS. THE IMPORTANT THING IS WHEN WE LOOK AT TISSUE MICRO RAYS AT LEVELS OF H2AZ, WE SAW THAT HIGH LEVELS OF T2AZ ARE VERY STRONGLY ME DICATIVE OF NEGATIVE OUTCOME WHEREAS LOW LEVELS ARE MEDIUM LEVELS AND THESE HAVE BEEN SCORED IN BY THREE INDEPENDENT PATHOLOGISTS TO CREATE THESE SCORES THAT WE USE IN THESE MINOR PLOTS. YOU GET MUCH BETTER OUTCOME. THIS IS AS GOOD A DIFFERENTIATION AS YOU SEE IN SOME LIKE THE EARLIER PLOT THAT I SHOWED OR BASED ON ERPR2 STATES. WE'VE GONE ON BECAUSE OF H2AZ BEING A CHROMATIN PROTEIN. H2AZ COMES IN AND DECORATES AROUND ACTIVE TRANSCRIPTIONAL START SITES AND REPLACES REGULAR OLD H2A IN THESE ACTIVE GENES. WE TOOK THE APPROACH OF SCREENING THROUGH LIBRARIES OF NOVEL HDAX INHIBITORS. WE FOUND A NUMBER OF LEAD MOLECULES WITH THE HELP OF OUR CELLULAR SCREENING FACILITY AS ONE OF THE CORES IN IGSP THAT ARE ACTING AT UNDER ONE MICRO MOLAR LEVELS. I WANT TO GO INTO A LITTLE BIT MORE DETAIL IS HOW EVERYBODY'S BEEN FOCUSING ON A HANDFUL OF WELL-KNOWN NUCLEAR RECEPTORS THAT ARE ASSOCIATED WITH BREAST CANCER AND BREAST CANCER OUTCOMES. BUT ACTUALLY, THERE ARE 48 NUCLEAR RECEPTORS THAT ARE ENCODED IN OUR GENOMES. AND OF THOSE 48 NUCLEAR RECEPTORS, TWO DOZEN OF THEM ARE EXPRESSED IN THIS MCF7 CELL LINE MODEL THAT WE'RE USING. IF YOU LOOK ACROSS MULTIPLE CELL LINES YOU GET MAYBE ANOTHER HALF DOZEN OR SO NUCLEAR RECEPTORS. AND POTENTIALLY ALL OF THESE NUCLEAR RECEPTORS OR ANY OF THESE NUCLEAR RECEPTORS MIGHT BE FAIR GAME. AND OF COURSE FOR MANY OF THEM YOU CAN PULL PHARMACEUTICALS RIGHT OFF THE SHELF. PHARMACEUTICALS INDUSTRY RELIES HEAVILY ABOUT 10-15% OF ALL DRUGS ON THE MARKET ARE TARGETED AGAINST THESE MOLECULES. SO THE PROBLEM WITH THIS IS THAT WE NEED IT TO DEVELOP SOME KIND OF TECHNOLOGY THAT OVERCAME THE LIMITATIONS OF ANTIBODIES BECAUSE IT'S JUST NOT POSSIBLE TO SYSTEMATICALLY GET ANTIBODIES TO ALL OF THESE TRANSCRIPTION FACTORS. SO THIS PROJECT HAS NOW BECOME PART OF THE HUMAN N-CODE PROJECT. I'M SHOWING YOU WHAT THE SORT OF SOME OF THE RESULTS THAT LED US TO THE INCORPORATION INTO HUMAN N-CODE WHERE THEY BASICALLY ARE AIMING TO GO AFTER AS MANY OR ALL OF THE TRANSCRIPTION FACTORS AND A FEW CELL LINE. WE'RE JUST INTERESTED IN THE NUCLEAR RECEPTORS. SO WE HAVE A -- APPROACH WHERE WE SIMPLY KNOCK IN A CASSETTE. IN THIS CASE WE'RE USING GFP, AND THEN WE CAN STABLY TRANSECT THE CELL LINES AND DO CHIP SEEK ON THOSE STABLY TRANSCRIPTION STABILIZE. THIS IS JUST SHOWING THAT TIME AFTER TIME, AND WE'VE MADE SOMETHING LIKE 150 DIFFERENT LINES NOW, 80 TO 90% OF THESE LINES ARE SHOWING THE APPROPRIATE LOCALIZATION FOR, I DON'T SHOW THIS ON THIS SLIDE BUT IF YOU LOOK AT VITAMIN D RECEPTOR FOR EXAMPLE YOU CAN SEE APPROPRIATE TRANSLOCATIONS. IT'S BINDING TO THE RIGHT SITES IN THE GENOME. THIS IS FOX A WHERE WE'VE TAKEN AN ANTIBODY AGAINST KNOX A ONE OR ANTIBODY AGAINST GFP AND YOU GET HIGHLY OVERLAPPING PROFILES ON THE ORDER OF 80-90% WHICH IS SIMILAR TO WHAT IF ON A GOOD DAY, IF BARBARA DOES A CHIP SEEK IN HER LAB IN THE N-CODE PROJECT AND I DO ONE OR MIKE SCHNEIDER DOES ONE IN HIS LAB WE SEE OVERLAP FOR THE SAME TRANSCRIPTION FACTOR FOR THE SAME CELL LINE. SO THIS WE FEEL IS PRETTY GOOD. SO WE'VE MAPPED OVER 40 FACTORS IN THIS PARTICULAR CELL LINE. AND THIS IS JUST SHOWING YOU THE RELATIVE DISTRIBUTION AROUND TRANSCRIPTIONAL START SITES. BASICALLY SHOWING YOU THIS SLIDE AS AN OVERVIEW OF ALL THE FACTORS THAT WE'VE MAPPED BUT ALSO BECAUSE THE FACTORS OF MAPS SO FAR THE MOST ARE ESTROGEN RECEPTOR AND ANDROGEN RECEPTOR. AND BASED ON THAT DATA, THERE'S THIS IDEA THAT AS YOU CAN SEE, ESTROGEN RECEPTORS IS SPREAD OUT RELATIVE TO TRANSCRIPTIONAL START SITE ALL OVER THE PLACE. SO THIS IDEA THAT THESE NUCLEAR RECEPTORS ARE ACTING THROUGH LONG-RANGE ENHARNSES T ENHANCERS TO THEIR TARGET GENES. THIS IS JUST TO DISPEL THE NOTION THAT THAT'S THE WAY NUCLEAR RECEPTORS WORK SOMEHOW. BECAUSE YOU ACTUALLY SEE THE WHOLE RANGE. YOU SEE SOMETHING LIKE TR2 OR YOU SEE NEURO 1, LXR. THESE ARE ALL VERY TIGHTLY ACCUMULATED AROUND THE TRANSCRIPTIONAL START SITE. SO PROBABLY NUCLEAR RECEPTORS ARE DEPLOYING A RANGE OF MODALITIES IN TERMS OF THE WAY THEY'RE ACTING ON THEIR TARGET PROMOTER GENES. >> KEVIN, BEFORE YOU MOVE ON, ARE THOSE UNDER THE INFLUENCE OF [INDISCERNIBLE] >> CORRECT. YES. SO IN THE CASES WHERE THERE'S A KNOWN LIGAND WE'VE ADDED THE LEGGEN IN THIS PARTICULAR EXPERIMENT. WE DON'T HAVE ALL OF THE DATA FOR THAT YET WHICH IS A WHOLE OTHER INTERESTING BUG. WE ALSO SEE HIGH LEVELS OF CONSERVATION AND WE HAVE A NICE FIGURE OF THIS IN THE 2009 CELL PAPER WHERE WE SHOW THAT YOU GET EVEN AT THE LEVEL OF NUCLEOTIDE TO NUCLEOTIDE VERY NICE PROFILE OF CONSERVATION ACROSS THE HALF SITES OF NUCLEAR FOR BINDING. WE DISCOVERED THESE HIGH SPOTS OF TRANSCRIPTION FACTOR BINDING. WE INITIALLY IDENTIFIED THESE IN DROSOPHILA AND THEN THAT CAUSED US TO LOOK IN THE HUMAN N-CODE PROJECTS HAVE BEEN LOOKING AT THIS AS WELL HOT SPOTS OF TRANSCRIPTION BINDING IN THE HUMAN GENOME. THAT IS BINDING OF MULTIPLE FACTORS WHERE YOU'RE SEEING MANY MORE FACTORS BINDING THAN YOU WOULD EXPECT UNDER SOME KIND OF RANDOM MODEL. AND THIS IS SUCH A RANDOM MODEL, IF YOU HAVE N FACTORS WITH Y SITES, SAY, YOU CAN RUN SIMULATIONS ABOUT THE DISTRIBUTION THAT YOU WOULD EXPECT. AND THIS IS ACTUALLY OBSERVED DISTRIBUTION. YOU SEE THAT AROUND EIGHT. YOU GET A HIGHER THAN EXPECTED AMOUNT. AND SO WE GENERALLY ARE CONSIDERING BOTH IN THE N-CODE PROJECT AND IN OUR HUMAN DATA EIGHT TO TEN IS THE AREA WHERE THINGS START TO GET HOT. HIGHLY OCCUPIED TARGETS IS ANOTHER ACRONYM FOR THAT'S BEEN MADE UP TO FIT THE IDEA OF HOT SPOTS. SO THE QUESTION IS WHAT'S GOING ON AT THESE HOT SPOTS. HERE WE SEE THAT THEY'RE ASSOCIATED WITH OPEN AND ACTIVE CHROMATIN. IN ADDITION TO THE TRANSCRIPTION FACTOR BINDING WITH MAP WE'VE DONE FAIR. WE'VE DONE A VARIETY OF MARKS, AND WHAT YOU CAN SEE IS THAT ALL OF THE MARKS THAT ARE ASSOCIATED WITH TRANSCRIPTION LIKE HVK4 METHYLATION, TR TRY TRY TRI METHYLATION EVEN A MONO METHYLATION WHICH IS MORE OF A MARK OF ENHANCERS. H3R17 METHYLATION ASSOCIATED WITH KARM2. BUT THEN YOU'VE GOT THE REPRESSIVE MARK WHICH DOESN'T SHOW THIS ASSOCIATION. SO OPEN CHROMATIN IS ASSOCIATION. WHAT'S NEAT ABOUT WHAT YOU CAN DO WITH THESE NUCLEAR RESOMPLET MODELS IS -- RECEPTOR MODELS IS YOU CAN VERY EXQUISITELY CONTROL WHEN YOU ACTIVATE GENES BY SIMPLY STARVING THE CELLS AND ADDING LIGAND. THAT'S WHAT WE'VE DONE HERE AND YOU CAN SEE THAT THE HOTTER A REGULATORY, THE MORE LIKELY THAT WHEN YOU ADDIS GEN ADD ESTROGEN AND WE'VE DONE THIS AS WELL THE MORE LIKELY YOU'LL HAVE A STRONG RESPONSE. THE TRANSCRIPTIONAL RESPONSE IS ACTUALLY ASSOCIATED WITH LEVEL OF BINDING OF NUMBER OF FACTORS THAT ARE ACCUMULATING HERE. THIS IS SORT OF INTERESTING FINDING FROM ALL OF THIS. WE'RE TRYING TO BUILD STRATIFYING SUBSETS OF DISEASE. ONE OF THE THINGS YOU CAN DO WITH THIS DATA IS CLUSTER IT ON BINDING SITE INSTEAD OF GENE EXPRESSION PATTERN. THAT IS AN EXAMPLE WHERE WE FOUND AN RECEPTOR AND -- RECEPTORS ARE IN THE SAME SET OF NODES CLUSTERING TOGETHER. WE WONDERED WHETHER THAT MIGHT BE FUNCTIONALLY SIGNIFICANT. WE TREATED CELLS BOTH WITH ESTROGEN AS WELL AS WITH A VARIETY OF EITHER COLLECTIVE OR PAN AGONIST FOR RENTNOIC. FOR 3W- 80% OF THE GENES THAT ARE TARGETED BY ESTROGEN RECEPTOR AND -- ACID RECEPTORS IS THAT THE ONES THAT ARE JUST TARGETED WITH THE RETNOIC ACID, THEY'RE HAVING THE OPPOSITE EFFECT OF HE IS GEN LYING ENGINES. LYING -- ESTROGEN LIGANDS. SO THIS SUGGESTED ANTAGONIST PARTICULAR EFFECT. WE WENT ON TO SHOW WITH SOME BIOCHEMICAL EXPERIENCE THAT THERE'S SOME DEGREE OF COMPETITIVE BINDING AT ABOUT 70% BUT THERE'S ALSO LONGER-RANGE ACTING EFFECTS HAPPENING FAT WE DON'T UNDERSTAND AS WELL. I WON'T SHOW THAT DATA BECAUSE I WANT TO SHOW YOU THIS DATA. THIS IS WHERE NOW YOU'VE GOT 300 TUMOR-EXPRESSION PROFILE IN THIS DATA SET OVER HERE. AND WE'VE OVERLAID ON THAT NOW BINDING PROFILES OF THESE TWO NUCLEAR RECEPTORS. AND WE EXTRACTED THE DIFFERENT SETS OF GENES NOW THAT HAVE DIFFERENTIAL EXPRESSION AND ALSO BINDING TO VARIOUS DEGREES. AND THEN WE REPLOTTED THE SURVIVAL CURVES OF THESE PATIENTS BASED ON THOSE SUBSETS. WHAT YOU CAN SEE HERE IS THAT IF YOU HAVE HIGH INDEX OF ESTROGEN ACTIVATED SITES, YOU HAVE A VERY HIGH PROBABILITY OF SURVIVAL. I'M SORRY OF RETNOIC ACID YOU HAVE HIGH PROBABILITY OF SURVIVAL. IF YOU SEE HIGH ESTROGEN SITES YOU HAVE THE OPPOSITE AND YOU SEE A MIDDLE GROUP AS WELL. THIS IS ALSO TRUE FOR RECURRENCE. SO WHAT THIS SUGGESTS ARE SUBSETS OF PATIENTS THAT MIGHT BENEFIT FROM RETNOIC ACID THERAPY WHICH HAS BEEN TRIED BUT NOT IN A WAY THAT'S BEING GUIDED BY HAVING MOLECULAR PROFILE THAT SAYS HEY THIS SET OF PATIENTS MIGHT ACTUALLY BE MORE LIKELY TO RESPOND THAN THIS OTHER SET OF PATIENTS. I THINK ANOTHER THING THAT'S NEEDED ARE BETTER RETNOIC DERIVATIVES THAT ARE LESS TOXIC. SO OKAY. SO ANOTHER THING THAT'S COME FROM THIS DATA IS ACTUALLY IT WAS A BIT SURPRISING IS WE SAW THAT PPR DELTA HAS DIRECT TARGET OR MAYBE IT'S NOT SURPRISING TO SOME PEOPLE. BUT IT JUMPED OUT TO US. ITS DIRECT TARGETS SIMILARLY LOOKED AT HAVE AN OPPOSITE EFFECT. THAT IS IF YOU HAVE HIGH LEVELS. THIS IS JUST PPR DELTA EXPRESSION LEVELS. THE HIGHER THE LEVELS OF EXPRESSION, ARE THE LOWER THE SURVIVAL. AND THE SAME WITH THE TARGET SCORES. SO THIS IS CORRELATING, YOU CAN READ THIS AT THE LEVEL OF JUST LOOKING AT PPR DELTA OR YOU CAN LOOK AT THE TARGET GENES OF PPR DELTA. AND GET SIMILAR KIND OF RESULTS. SO THIS WOULD SUGGEST THAT BY GIVING PPR DELTA AGONISTS OR ANTAGONISTS THAT YOU MIGHT IMPROVE. AND SO THERE ARE ACTUALLY A COUPLE DRUGS THAT HAVE BEEN DEVELOPED AS PPR DELTA ANTAGONISTS, BUT THEY'RE NOT IN WIDE USE. SO IN THE LAST COUPLE MINUTES, I WANT TO TALK ABOUT THE LAST THING I PROMISED WHICH WAS SORT OF HOLY GRAIL IS BEING ABLE TO STRATIFY NOT UNCELLULAR STATE. AND THIS IS SORT OF WHERE WE STARTED. YOU CAN ADD THE SINGLE MOLECULES THAT YOU WANT TO ULTIMATELY PAIR TO THERAPEUTICS. YOU CAN USE COMMONTORIC TO BETTER STRIDE FI. IT'S BETTER TO BUILD DIAGNOSTIC TESTS AROUND THE GENETIC INFORMATION THAT'S AFFECTING THE CELLULAR STATE THAT YOU'RE MEASURING IN ALL THIS. SO WE DID A VERY, THIS IS JUST A PILOT STUDY AND WE'RE TRYING TO CONFIRM IT WITH TCDA DATA RIGHT NOW. BUT I'LL WALK YOU THROUGH IT VERY QUICKLY. WE LOOKED AT ABOUT TWO DOZEN CANCERS. ER POSITIVE AND TRIPLE NEGATIVE. WE DID RNA SEEK IN THIS STUDY. WHEN YOU DO RNA SEEK YOU SEE A CIVIL HER THING AS PCA1, PCA 26789 I MEAN SORRY PCA ANALYSIS, PRINCIPAL COMPONENTS ANALYSIS IN THE FIRST TWO COMPONENTS. YOU SEE THE SIMILAR THING AS IF YOU DID MICRO RAY EXPRESSION WHERE YOU SEE SOME STRATIFICATION BASED ON EXPRESSION LEVELS. WE CAN USE ISOFORMS. I'M JUST GOING TO BLOW THROUGH THIS TO INCREASE THAT STRATIFICATION. BUT ULTIMATELY THE QUESTION IS CAN YOU USE VARIANTS THAT IDENTIFIED ARE FROM THIS SUBSET. AND SO WE CHOSE TO LOOK AT RNA SEEK BECAUSE THIS IS FIRST IT WAS LESS EXPENSIVE AND IN DOING WHOLE GENOME OR EVEN EXOME, PARTICULARLY WHEN WE SAWRT STARTED THIS. IT'S FOUND TO GO MOST RELEVANT BECAUSE IT'S THE EXPRESSED VARIATION. FROM THESE PATIENTS TUMORS WE FOUND GREATER THAN 35,000 SINGLE NUCLEOTIDE VARIANTS. WE DON'T KNOW WHETHER THEY ARE SCHEMATIC OR GERMLINE. WE LOOK AT ALLELES ARE THEY RARE VERSUS COMMON. I CAN GO INTO MORE DEALS HOW WE DEFINE OR WE USED REVOLUTIONARY CONSTRAINT USING 30 VERTEBRATES OR MAMMALS IN SEQUENCE NOW NUCLEOTIDE BY NUCLEOTIDE SCORES. WHAT YOU CAN SEE IS AS YOU MIGHT EXPECT A RARE VARIANTS ARE SKEWED TO BE MORE DELETERIOUS. THOSE ARE MORE RECENT AW ADDITIONS TO THE GENOMES. IF YOU LOOK AT THE COMMON VARIANTS, WHAT YOU SEE IS AT UNIVERSITY OF CHICAGO, WE JUST SELECTED 24 MEASURES AT RANDOM OR 25 -- PATIENTS AT RANDOM, OR 25. AND WE HAVE A MIX, FAIRLY EVEN MIX OF CAUCASIAN AND AFRICAN AMERICAN PATIENTS. AND YOU CAN SEE THAT IF YOU LOOK AT THE COMMON VARIANTS JUST LIKE G-WIDE STUDIES NO PROBLEM, WE CAN GRATIFY BASED ON GRACE. SO THIS DOESN'T HELP US STRATIFYING BASED ON DISEASE, RIGHT? SO THIS IS WHAT, SORT OF STUNNED US. WHEN WE LOOK AT THE RARE AND DELETERIOUS VARIANTS, SUDDENLY IT DOESN'T MATTER THE RATIO BACKGROUND, WE'RE FINDING STRATIFICATION OF TRIPLE NEGATIVE AND ER POSITIVE BASED ON BEING RARE AND DELETERIOUS. AND WE WONDERED, IS THIS GERMLINE OR SCHEMATIC S SMAT SOMATIC. IF THIS CAN BE VALIDATED, ONE COULD GET YOUR GERMLINE GENOME DONE AND ANALYZE IT IN SUCH A SYSTEM AND FIGURE OUT WHETHER FOR EXAMPLE YOU SHOULD BE SCREENING WHEN YOU'RE 30 OR YOU SHOULD BE SCREENING WHEN YOU'RE 50. IF YOU DO FIND SOMETHING, WHETHER GO AFTER AGGRESSIVELY OR NOT SO AGGRESSIVELY. IT BRINGS OUT THESE RARE VARIANTS FALL AND IN AING MOLECULAR PATHWAY THE MEANING OF WHICH ARE WELL-KNOWN TO BE ASSOCIATED WITH CANCERS. WE CAN DO, IF YOU'RE A POPULATION GENETICISTS, YOU LIKE TO SEE FST TESTS TO SHOW YOU'VE GOT TWO DIFFERENT POPULATIONS. I'LL JUST DRAW YOUR ATTENTION TO THE RIGHT HERE. THE RIGHT IS LITTL LOTS OF DIFFERENT REPETITIONS WHAT WE DID ON THE LEFT. WE CALCULATED SFP IN THE SIMULATED POPULATION WITH THE SAME DATA BUT JUST RANDOMIZING WHETHER YOU'RE IN ONE BILLION OR THE OTHER, TRIPLE NEGATIVE OR ER POSITIVE IN QUOTATION MARKS. AND THIS IS OBSERVED AT SCORES OF FOUR AND FIVE. SO HERE IS A WHOLE CONTINUUM AND YOU CAN SEE IF YOU'RE COMMON, IF WE DO THAT WITH COMMON VARIANTS, EVEN IF WE USE COMMON DELETERIOUS PREDICTABLY DELETERIOUS WE DON'T GET ANY SIGNAL. BUT IF WE USE THE RARE VARIANCE, THAT'S WHEN THE SIGNAL BREAKS UP. SO THIS SFP IS GREAT AS IF YOU WOULD SEE WITH COMMON VARIATIONS BETWEEN OF COMES, ARBIANS AND CHINESE. AS A FORMER POPULAR GENETICIST THIS IS LIKE FALLING OFF A TRUCK. SO WE'RE EXCITED ABOUT THIS. WE'RE CURRENTLY CHURNING OVER THE CPTA GOT. WE'RE HAVING TO RECALL ALL THE BASIS AND STUFF. WE'RE PART OF THE ANALYSIS WORKING GROUP FOR BREAST CANCER AND WE'RE NORMALLY NOT DOING THIS KIND OF ANALYSIS THAT WE NEED SO WE'RE RECHURNING OVER THIS SO WE HAVE A HUNDRED CORES OR SO RUNNING THE TCPGA DATA RIGHT NOW TO SEE IF WE CAN VALIDATE. I DON'T WANT TO LEAVVALIDATE. I WANT TO LEAVE YOU WITH THESE LAST CONCLUSIONS. EQUILIBRIUM ALONG WITH RARE VARIANCE MIGHT MAKE THE TRADITIONAL GWAS AN UPHILL BATTLE. WE CAN TALK ABOUT WHY IN MORE DETAIL. AND PERHAPS OTHER CANCERS AND MAYBE OTHER HUMAN DISEASES ARE BASED IN THE FUNDMENTAL RULES. SO MY TIME IS JUST ABOUT UP. I DID THROW IN ONE OTHER STUDY THAT I TOLD YOU I MAY NOT GET TOO. I'LL GIVE YOU THE TAKE HOME. SO IN THE FIRST STUDY I TALKED ABOUT, WE WENT FROM BUILDING NETWORKS IN FLIES TO FINDING A NEW GENE IMPLICATED IN A HUMAN CANCER. IN THIS STUDY, WE'VE GONE FROM LOOKING AT CANCER GENOME GOT TO USING FLIES TO TEST FOR NEW TUMOR SUPPRESSOR. AND SPECIFICALLY FOR. YEARS IT'S BEEN KNOWN THAT MONO SOMY SEVEN IN AL'S IS HIGHLY ASSOCIATED WITH NOT ONLY ML'S BUT ALSO OTHER MYELOID DISEASES. MICHELLE AND OTHERS IN THE FIELD, MICHELLE IS A COLLEAGUE OF MINE AT THE UNIVERSITY OF CHICAGO HAD MAPPED THE MINIMAL AREA. THEY ACTUALLY DELETED THE SENT TENIC REGION IN MICE AND DID NOT SEE A PHENOTYPE. IT TURNED OUT THAT BY DOING COPY NUMBER MAPPING, WE MAPPED MINIMAL REGION IN 35 INDIVIDUALS THAT WE LOOKED AT BUT WE ALSO LOOKED AT GENE EXPRESSION AND WE ASKED ARE YOU AWARE OF THE SUFFICIENT, INSUFFICIENT SORT OF LOW TIDE THE LEVEL OF GENE EXPRTION. AND WE FURTHER NARROWED IT. AND THEN WE GOT LUCKY AND WE FOUND IN ONE INDIVIDUAL WE FOUND A SMALLER REARRANGEMENT THAT BASICALLY BROKE UP THE COX 1 GENE WHICH WAS PROXIMAL TO THE REGION THAT DIDN'T SHOW A PHENOTYPE IN MICE BUT HAD BEEN MAPPED BY SORT OF TRADITIONAL METHOD BEFORE HOPEFULLY EXPLAINING WHY THE MUST STUDY DIDN'T LEAD TO A TUMOR SUPPRESSOR. ATE TURNSUPPRESSOR. IT TURNS OUT COX 1 HAS DIFFERENT LEVELS OF DOSAGE AND DIFFERENT LINEAGES AND SO ITS REGULATION SEEMS TO BE FINALLY TUNED. DURING DIFFERENTIATION A MYELOID LINEAGE, WE CONFIRM IN PRIMARY TUMORS AND THEN WE MOVED INTO THE FLY MODEL WHERE WE CAN VERY SPECIFICALLY KNOCK DOWN THE COX 1 ORTHO LOG CUT WHICH WAS A TRANSCRIPTION FACTOR. BACK TO THE HOMEODOMAIN. WHAT WE GET BETWEEN 20 AND 30% WE HAVE TWO DIFFERENT KNOCK DOWNS. SO IT'S BEEN REPLICATED WITH TWO DIFFERENT KNOCK DOWN CONSTRUCTS. WE GET THESE MELON OTIC TUMORS. SKIPPING TO THIS IT'S HARD TO SEE IN THESE IMAGES ON THE LEFT BUT IF YOU LOOK TO YOUR RIGHT, YOU CAN SEE THAT WHEN WE ACTUALLY COUNT BLOOD CELLS. WE'RE ONLY KNOCKING THESE DOWN IN THE BLOOD. YOU GET HYPER PROLIFERATION. SO WE'RE BASICALLY GIVING THESE LITTLE FLIES BUG CANCER. OKAY. I'M GOING TO SKIP ALL THIS. SOME OF YOU WANT TO TALK TO ME ABOUT DATA CLOUDS. WE'VE BUILT ONE. WE'RE USING IT. IT'S TIED IN TO A VERY BIG NETWORK OF HIGH SPEED INTERNET. WE'RE TRYING TO MAKE IT EASY FOR PEOPLE TO USE. A LOT OF PEOPLE ARE USING IT AND A LOT OF PAPERS ARE COMING OUT THAT HAVE BEEN BUILT ON IT. THE N CODE ARE GOING INTO IT. IT'S CALLED BIONIMBUS. SEND ME AN E-MAIL IF YOU WANT TO HEAR ABOUT IT. IT'S PART OF A SORT OF BIGGER VISION WHERE WE WOULD BE PULLING IN BOTH CLINICAL DATA LAYERING THAT ON TOP OF ALL THE GENOMIC DATA IN THIS CLOUD AND USING IT TO DRIVE DEVELOPMENT OF BOTH MODELS, MODEL SYSTEMS IN THERAPEUTIC AND DIAGNOSTIC DEVELOPMENTS. I NEED TO THANK THE PEOPLE WHO ARE PART OF OUR INTERDISCIPLINARY CANCER GENOMES PROJECT AND THE PEOPLE IN BLUE ARE PART OF THE SPOP PROJECT AND THE PEOPLE IN GREEN PART OF THE NUCLEAR RECEPTOR PROJECT. THANK YOU. [APPLAUSE] >> PEOPLE THAT MEAD TO BREAK K ARE THERE QUESTIONS? >> I JUST WANTED TO UNDERSTAND A LITTLE BIT MORE OF YOUR IMPLICATION ABOUT THE GWAS AND BREAST CANCER. IS IT BECAUSE THE GWAS HAS BEEN DONE IN WAY THAT HASN'T FIRST SUBSETTED BREAST CANCER INTO DEFINED DISEASE TYPES AND THAT IS THE LACK OF SIGNAL, OR IS IT THAT THE THINGS THAT YOU'RE FINDING TO BE ASSOCIATED ARE NUMBER ONE RARE AND HAVE TO BE FILTERED BASED ON THERAPEUTIC FUNCTIONS CHANGING AMINO ACIDS. >> NO MORE, SO I THINK THOSE ARE TWOIVTWO DIFFERENT DISCUSSIONS. WHAT I'M ARGUING TODAY PUBLICLY IS THAT IT'S MORE OF THE LATTER. AND THAT THE SORT OF TECHNICAL PROBLEM IS THAT IF THE CAUSATIVE VARIATION THAT SORT OF SETS YOU UP JETALLY T GENETICALLY TO BE PREDISPOSED FOR THIS IS RARE AND ALSO IT HAS TO HAVE THE TRADITIONAL CONDITION THAT MAKES IT ESPECIALLY RARE. THEN THE PROBABILITY OF HAVING A COMMON ALLELE THAT'S ON A HALF A TYPE BLOCK THAT CONTAINS THAT VARIANT. BECAUSE WE ALL HAVE OUR OWN PERSONAL VARIANTS, OKAY. IT'S VERY LOW. AND SO EVEN THOUGH YOU'RE MEASURING THAT ALLELE THAT IS IN LINKAGE TO SOME SUBSET. LET'S SAY IT OF IT'S ONLY 1%. SO LET'S SAY THAT ALLELE IS ONLY IN LINKAGE IN 1% OF ALL REPRESENTATIONS OF THAT ALLELE THAT YOU'VE MEASURED. IT MAKES IT LIKE STATISTICALLY IMPOSSIBLE TO DETECT THE REAL EFFECT THOSE RARE VARIANTS. >> BUT THE VARIANTS THEMSELVES WERE ALREADY IDENTIFIED AS SNIPS IN THE HUMAN POPULATION, RIGHT? >> NO. ONLY IN THE GWAS STUDIES. >> NO, BUT I MEAN THE RARE ONES. >> IN OUR STUDIES, THESE RARE ONES ARE ONES THAT DIDN'T APPEAR IN HA p38MAPK THEY DON'T APPEAR IN THE THOUSAND GENOMES PROJECT. SO IF YOU POSIT THAT AND WE KNOW THIS THAT WE'RE CARRYING OUR OWN SORT OF FAMILY IN RECENT HISTORY OF MUTATIONS. SOME OF THOSE ARE DELETERIOUS. YOU CAN IMAGINE THAT IF YOU HAVE A VALUATOR -- ONE OF THE MORE COMMON VARIANTS THAT WE FOUND WAS IN THREE PRIME UTR OF MYC. IT HAPPENS TO FALL RIGHT OVER A SUITOR THAT'S BEEN VERIFIED AS A MICRO RNA TARGET. AND SO IMAGINE THAT WE, YOU KNOW, THAT THE THAT AREA GETS MUTATED. THOSE COUPLE AMINO ACIDS -- I MEAN NUCLEOTIDES, GET MUTATED, A BUNCH OF DIFFERENT TIMES. BUT ON DIFFERENT HAPPE HAPLOTIDES . IT HAPPENS IN THE GWAS STUDIES. I THINK THAT'S A BIG OPEN QUESTION. AS WE DIG DEEPER AS WE GO INTO EXOME SEQUENCING AND GENOME SEQUENCING, AND WE START TO BE ABLE TO FOR THE FIRST TIME LOOK AT THESE RARE ALLELES AND THEN PUT ADDITIONAL FILTERS OVER THE TOP OF THEM LIKE CONCENTRATION. WE DID SOMETHING VERY SIMPLE HERE, CONCENTRATION. BUT YOU CAN IMAGINE TAKING THESE NETWORKS AND WE'RE IMAGINING THAT. TAKING THESE NETWORKS THAT WE'VE BUILT WITH THE NUCLEORECEPTORS, WITH THE GENE EXPRESSION PROFILES AND SO FORTH AND FILTERING THROUGH THE GENOME DATA THROUGH THAT. YOU MAY END UP WITH SOME VERY POWERFUL TOOLS FOR MAKING PREDICTIONS AT LEAST FOR THOSE THINGS THAT HAVE THE STRONG GENETIC COMPONENT. >> SO I REALLY LIKE THE PCA STUDY THAT YOU DID AND FOUND THE DIFFERENT RACES BUT NOT THE DISEASE STRATIFICATION. I'M CURIOUS INSTEAD OF USING ANOTHER SUBSET TO FIND THE DISEASE STRATIFICATION, IF YOU LOOKED AT OTHER PRINCIPAL COMPONENTS OR IF YOU HAD JUST TAKEN ONE OF THE POPULATIONS AND DONE PCA ON THAT, WOULD YOU HAVE RECOVERED THE DISEASE STRATIFICATION IN THAT MANNER? >> YES. WE, AS WE DIG DOWN IN PRINCIPAL COMPONENTS, THERE'S JUST NOT ENOUGH SIGNAL TO GET THE DISEASE STRATIFICATION OUT OF THAT OUT OF THE COMMON VARIANTS. AND PART OF THAT AGAIN MEANS I GO BACK TO THE DISCUSSION JUST HAVING WITH LOU ABOUT, IF IT IS RARE DELETERIOUS VARIANTS THAT ARE IMPORTANT, THEN YOU'RE JUST NOT GOING TO, YOU'RE JUST NOT GOING TO HAVE THOSE ASSOCIATIONS THERE IN THE FIRST PLACE. >> SO YOU GET ABOUT 80% OF THE VARIANTS FROM THE FIRST TWO PRINCIPAL COMPONENTS. >> I FORGET. >> JUST CURIOUS. THANKS. >> IF YOU SEND ME AN E-MAIL I CAN TELL YOU EXACTLY WHAT PERCENT OF THE VARIANTS IS EXPLAINED BY THE FIRST TWO COMPONENTS. >> YOU MENTIONED THAT THE TOTAL -- TENDED TO DEPEND ON THE NUMBER OF [INDISCERNIBLE] WHICH WAS BINDING TO A GIVEN LESION. ARE THESE TRANSCRIPTIONAL FACTOR OR THE BINDING SITE CORRESPONDING TO THEM FUNCTIONALLY RELATED TO EACH OF THEM SUCH THAT THEY'RE DRIVING A CELL IN A PHENOTYPIC DIRECTION AND NOT GOING EVERYWHERE? >> I WOULD ARGUE THAT AT SOME LEVEL THEY MUST BE BECAUSE WE CAN USE THEM TO STRATIFY DIFFERENT SUBPOPULATIONS OF PATIENTS THAT ARE, HAVE DIFFERENT SURVIVALS AND DIFFERENT LEVEL BIOLOGICAL DEGREES OF RESPONSE TO THERAPY AND THEREFORE SURVIVAL. >> SO YOU DON'T KNOW AT THIS STAGE THE TRANSCRIPTIONAL FACTORS HAPPEN TO CONNECT BY SOME SIGNA CENTRAL NODE, THEY DON'T FALL INTO A NETWORK. >> THEY DO. I DIDN'T SHOW YOU THE NETWORK DIAGRAMS. I FIGURED I WOULD PROBABLY ALREADY OVERDONE. I THINK THAT IN ANY ONE TALK, ONE GETS A LIMITATION OF NETWORK DIAGRAMS. IF YOU PASS THAT LIMITATION THEN PEOPLE ACTUALLY START THROWING PENCILS AND BOOKS AND STUFF AT YOU. SO I'VE GOT LOTS OF NETWORK DIAGRAMS OF THE NUCLEAR RECEPTOR. AND IT'S INTERESTING, I FIND IT INTERESTING THAT WE CAN ACTUALLY SEE THAT THERE'S ONE NETWORK, ONE SUBNETWORK THAT HAS A HIGH DEGREE OF CONNECTIVITY ASSOCIATED WITH AMONG A BUNCH OF FACTORS THAT ARE AROUND RETNOIC RECEPTOR AND ESTROGEN RECEPTOR AND PR AND GR AND SO FORTH. THEN THERE'S ANOTHER ONE THAT'S MORE JUNE FOAA ORIENTED AND THAT ONE WE CALL THE PROMOTER OR TRANSCRIPTIONAL START SITE NETWORK WHEN YOU SEE ATF AND ALL OF THOSE THINGS. SO WE THINK THAT THE SORT OF PROXIMITY IN NETWORK SPACE, IF YOU WILL, IS CORRESPONDING TO SOME BIOLOGY. BUT IT MAY BE DIFFERENT BAWLINGS DEPENDING -- BIOLOGIES DEPENDING ON WHICH SPACE YOU'RE LOOKING AT. >> THANK YOU. >> I WAS ACTUALLY A BIT CONFUSED ABOUT HOW RARE THE SNIPS YOU USE TO CLASSIFY THE CANCERS ARE. YOU MENTIONED THAT THEY ARE -- NOR THE THOUSAND GENOME. >> RIGHT. >> BUT HOW MANY PEOPLE DID YOU SEQUENCE IN THIS CASE? >> I THINK WE DID 25. >> RIGHT. SO THE FACT THAT YOU HAD THE PCA CAN EXPLAIN MORE VARIANTS THAN EXPECTED IMPLIES THAT THERE'S SOME CORRELATION STRUCTURE TO THOSE SPRAIRNTS ACROSS THE SET OF VARIANTS THAT YOU USED, RIGHT? >> WELL, WHAT I NEGLECTED TO TELL YOU IS THE WAY WE CALCULATED OUR, THE WAY WE CALCULATED THINGS IS THAT IT'S AT A GENEIC LEVEL SO YOU CAN HAVE MULTIPLE RARE VARIANTS THAT ARE CONTRIBUTING TO A GENE'S EFFECT RATHER THAN AT HAVING A PARTICULAR VARIANT. ALTHOUGH WE DID FILTER IN SUCH A WAY THAT WE HAD TO HAVE AT LEAST TWO INDIVIDUALS THAT HAD A PARTICULAR VARIANT. >> I SEE. SO KIND OF COMMON IN YOUR POPULATION, IN YOUR VERY SMALL POPULATION BUT VERY RARE. >> ASSOCIATED WITH CANCER. >> GOOD, THANKS. >> I'M INTERESTED IN THE NETWORKS YOU'RE BUILDING WITH DIFFERENT LEVELS OF DATA. AND SO LITERATURE, DIFFERENT COMPONENTS AND ADDING DIFFERENT ENDLEDGES BASED ON THOSE AND I'M KIND OF, I'M DOING SOMETHING -- I'M KIND OF WONDERING HOW, WHERE CAN I DO THE INTERSECTION SO THE UNION OF ALL THE DIFFERENT DATA SO KIND OF SEEING HOW THINGS GO TOGETHER AND HOW THINGS BALANCE AGAINST EACH OTHER AND ESPECIALLY A LITERATURE YOU GET CONFLICTING INFORMATION. AND SO WHEN YOU, IT LOOKS LIKE MOSTLY YOU'RE GETTING INFORMATION FROM THAT BY USING SOME NETWORK THEORY CONNECTIVITY AND THINGS LIKE THAT. DO YOU WEIGH EVERYTHING THE SAME? BASICALLY HOW IS IT NOT JUST A BIG MESS AND HOW DO YOU MAKE SOMETHING USEFUL OUT OF IT. >> YES. SO THAT'S A VERY VERY LONG ANSWER TO THAT THAT REQUIRES BEER. BUT WE DID PUBLISH A PAPER WITH ANDRE AND ANDRE'S PUBLISHED A WHOLE SERIES OF PAPERS ABOUT TELING WITH SOME OF THESE ISSUES. THE ONE THAT WE SORT OF DEALT WITH AND THE PAPER THAT CAME OUT IN 2006 WAS CREATING A MODEL THAT EXPLICITLY MODELS FALSE POSITIVE AND FALSE NEGATIVE RATES, YES. TO TRY TO DEAL WITH THE FIRST ISSUE THAT YOU BROUGHT UP OF CONFLICTING RESULTS. IT KIND OF EVOLVES INTO A FUNNY PAPER BECAUSE WHAT HAPPENED WAS ENTREE CREATED THIS VERY NICE ELEGANT MODEL. AND WHAT HE FOUND AND SO WHAT WE DID WAS WE LOOKED AT THE LITERATURE. YOU SAY WELL, YOU SAY THAT A INTERACTS WITH B AND LOU SAYS A DOESN'T INTERACT WITH B AND I SAY IT INTERACTS WITH B. WE FOLLOW TEMPERATE PATTERNS AND WE BUILT A MODEL AROUND HOW SCIENTIFIC CONSENSUS BASIS CAME ABOUT AND HOW CONFLICT OF STATEMENTS AND SO FORTH. OF COURSE MADE ASSUMPTIONS. BUT IN THE END ANALYSIS, IT TURNED OUT THAT THE MODEL CONVERGED ON TWO TABL STABLE STATES. ONE STABLE STATE WAS THAT SOMETHING ON THE ORDER OF 90% OF ALL STATEMENTS IN THE LITERATURE WERE TRUE. AND THE OTHER STABLE STATE THAT 95% WERE FALSE. THEY ARE TWO LIKE YOU KNOW GLOBAL MINIMUM IN THIS MODEL. THERE WASN'T ANYTHING IN BETWEEN. AND ANDRE AND I BACK THEN FIVE YEARS OR SO, SIX YEARS AGO, WE WOULD GO AROUND AND BEFORE SHOWING THE RESULTS, ASK PEOPLE WHAT THEY THOUGHT IN THEIR FIELD WAS THE CASE. WAS IT MORE, WAS MOST OF WHAT PUBLISHED GUARD BUDG GARBAGE OR WAS IT RIGHT. IT WAS INTERESTING TO SEE IT WAS ABOUT 50/50. [LAUGHTER] THAT WAS A LITTLE SCIENTIFIC SURVEY. BUT YOU KNOW, ADDRESSING HOW DO YOU KEEP IT FROM BEING A BIG MESS, FOR US, KEEP IT SIMPLE, YOU KNOW. DO SIMPLE THINGS. GET YOUR POWER OUT OF DOING A SIMPLE NON-WEIGHTED COMBINATION OF PROBABILITIES RATHER THAN TRYING TO WEIGHT THEM. IF YOU FIND THAT ONE TYPE OF DATA IS NOT VERY USEFUL FOR YOU TO CONVERGE ON YOUR, ON THE HIGHEST PROBABILITY EVENTS, THEN DON'T WORRY ABOUT THAT. THROW IT OUT. FROM YOUR ANALYSIS. SO KEEP IT SIMPLE, DO WHAT WORKS. >> ONE MORE. >> THERE ARE TWO PAPERS, THERE ARE TWO PAPERS NOW SUGGESTING THAT EDITING MAY BE MORE PREVALENT THAN WE THOUGHT IN THE TRANSCRIPTION. DID YOU FIND THIS TO BE IN THE CASE OR WERE YOUR VARIANTS ENCODED WITH THE GENOMIC IN IT? >> SO OUR VARIANTS THAT WE TESTED WERE ALL ENCODED IN A GENOMIC DNA. I THINK I SHOULD 113. THEY DID MORE THAN THAT. SO WE DIDN'T LOOK FOR EVIDENCE OF RNA EDITING. SO WE DIDN'T FIND IT. SO THE REAL ISSUE IS IF YOU LOOK FOR RNA EDITING, CAN YOU FIND IT AS OPPOSED TO IF YOU LOOK AT A WHOLE BUNCH OF VARIANTS, WHAT PERCENTAGE, PROBABLY ONLY A SMALL PERCENT PROPORTIONALLY OF ALL VARIATION OF RNA VARIANT THAT YOU FIND IN RNA IS ACTUALLY DUE TO RNA EDITING IS MY GUESS. AND SO YOU HAVE TO LOOK FOR IT TO FIND IT. I'M NOT AN EXPERT ON RNA EDITING. SO MAYBE SOMEONE WOULD BEG TO DIFFER. >> AS ONE WHO HAS DONE AN RNA EDITING STUDIED [INDISCERNIBLE] AND THEN ASK HOW MUCH IS IN FOR EXAMPLE PROTEIN CODING SEQUENCE OR PERTINENT NON-REPEAT THREE PRIME UTRs NOT SO MUCH. IT'S THERE. BUT IT'S NOT A PERVASIVE TSUNAMI OF NEW VARIATION. >> OKAY.