>> I'M WARREN KIBBE, I'M AT THE NATIONAL CANCER INSTITUTE. IT'S MY PRIVILEGE TO WELCOME Y'ALL TO THE PiCo LIGHTNING TALKS PART OF Pi DAY. WE ALL REALIZE THIS IS NOT ACTUALLY Pi DAY AND BACK AT Pi DAY IF YOU RECALL THAT WAS THE ONE BIG SNOWSTORM THE D.C. AREA GOT SO THIS HAD TO GET RESCHEDULED. BEFORE I GO ON AND TALK ABOUT THE LIGHTNING TALKS I WANT TO THANK LISA FEDERER FOR ORGANIZING THIS, SHE'S DONE A GREAT JOB OF MAKING SURE THAT WE CAN MAKE THIS HAPPEN EVEN THOUGH IT'S NOT Pi DAY ANY MORE. I WANT TO TALK A LITTLE BIT ABOUT THE IMPORTANCE OF Pi DAY AND REALLY WHAT IT MEANS FOR NIH AND BIOMEDICAL RESEARCH IN GENERAL. THAT'S -- ONE OF THE IDEAS HAS BEEN PUT FORWARD IS WE WANT TO CREATE Pi PEOPLE. BY THAT I DON'T MEAN PEOPLE WHO EAT PIE, THAT WOULD BE OKAY AND NOTHING WRONG WITH THAT PARTICULARLY ON Pi DAY BUT PEOPLE WHO ARE TRAINED DEEPLY IN BIOMEDICAL RESEARCH AND DEEPLY IN SOMETHING THAT'S QUANTITATIVE. SO WHETHER YOU'RE A MATHEMATICIAN, ENGINEER, A PHYSICIST, A CHEMIST, AND BE ABLE TO THEN PERSONALLY STRADDLE THOSE TWO AREAS AND BRING THAT COMPUTATIONAL VIEW INTO BIOMEDICAL RESEARCH. THAT'S SOMETHING AGAIN WE HAVE BEEN TRYING TO ENCOURAGE FROM NOT JUST FROM A WORK FORCE DEVELOPMENT STANDPOINT BUT ALSO TRAINING STANDPOINT. I DON'T WANT TO BELABOR THAT POINT BUT AGAIN, I WOULD LOVE TO SEE HOW MANY PEEP VIEW THEMSELVES GIVEN MY DEFINITION AS Pi PEOPLE, THAT ARE HERE? AND HOW MANY OF YOU ARE PRESENTING OF THE Pi PEOPLE? YEAH. SO I THINK THAT THAT'S ONE OF THE THINGS YOU WILL SEE IS THAT MANY OF THE PRESENTERS ARE IN FACT ARE THESE -- THE FOLKS THAT ALREADY ARE DEEPLY STRADDLED COMPUTATION AND QUANTITATIVE SCIENCES AND THEN OF COURSE BIOMEDICAL RESEARCH. I GIVE A LITTLE BIT OF LOGISTICS, LISA WILL TALK TO ALL THE PRESENTERS ABOUT WHAT YOU NEED TO DO WHEN YOU COME UP HERE BUT OF COURSE THE IDEA OF IT BEING A PiCo TALK, WE WANT IT TO BE -- I THINK IS IT THREE SLIDES, ONE IDEA, IN FOUR MINUTES SO 3.14. I THINK WE'RE GOING TO TRY NOT TO ASK TOO MANY QUESTIONS IN THE MIDDLE, THERE WILL BE A FORUM AFTERWARD FOR QUESTIONS. WE'LL SEE HOW SMOOTHLY WE CAN MAKE THE TRANSITIONS HAPPEN AND IF THEY DON'T HAPPEN QUITE AS SMOOTH THAT'S A PERFECT TIME TO ASK QUESTIONS WHILE THE NEXT PERSON IS COMING UP. WITH THAT, LISA, DO YOU WANT TO COME UP HERE? >> THANK YOU. ALL RIGHT. SO PRESENTERS. YOUR SLIDES THAT YOU HAVE AWESOMELY SENT ME, ARE ALREADY AND IN THE ORDER THAT YOU'RE GOING TO BE PRESENTING SO WHEN YOU SEE YOUR SLIDE POP UP, COME UP AND GIVE YOUR TALK. THERE'S A LASER POINTER HERE YOU MAY USE. YOU HAVE ADS WARREN SAID, THREE SLIDES, ONE IDEA FOUR MINUTES. SO PLEASE KEEP TO YOUR ALLOTTED TIME SO WE MANAGE TO GET EVERYONE IN. THEN AFTER YOU'RE DONE WITH YOUR TALK ADVANCE THE SLIDE TO NEXT SPEAKER'S TALK. SO WE WILL BEGIN. WITH HAROLD. THANK YOU. >> THANK YOU. SO WE LIVE IN THE GREAT AGE OF MATH MAKING. I THINK SOME OF YOU ARE PROBABLY THINKING SKY IS 5 -- THIS GUY IS 500 YEARS TOO LATE TO THE PARTY WHEN WE FINISHED WITH THE CONTINENTS BUT THIS IS A NEW ERA OF TRYING TO MAP THE BRAIN. THAT'S A PRIMARY FOCUS OF PRESIDENT OBAMA'S BRAIN INITIATIVE, PRODUCING A DETAILED MAP OF ACTIVITY AND STRUCTURE OF THE HUMAN BRAIN. SO MY LAB IS DEEPLY INVOLVED WITH THIS. WE DON'T USE THE HUMAN BRAIN. WE USE THE ANIMAL THAT I HAVE UP HERE. THIS IS THE ZEBRAFISH. THE LARVA OR BABY ZEBRAFISH. SO WE USE IT BECAUSE IT HAS ONLY ABOUT A HUNDRED THOUSAND NEURONS AND IT'S POSSIBLE TO GO IN AND TRY TO UNDERSTAND THE FUNCTION OF EVERY NEURON IN THE BRAIN. WHAT'S CRITICAL IS THAT THOSE NEURONS ARE KIND OF LAID OUT IN ALMOST EXACTLY THE SAME PATTERN AS HUMAN BRAIN SO YOU CAN LOOK AT THE HUMAN BRAIN AND YOU CAN FIND VERY SIMILAR PARTS IN THE ZEBRAFISH BRAIN, GIVE US A CHANCE TO USE THE SIMPLE SYSTEM TO IDENTIFY FUNCTIONING PARTS OF THE HUMAN BRAIN. BUT WE FACED A HUGE PROBLEM. THE PROBLEM IS THAT REMARKABLY WE DON'T HAVE A GOOD MAP OF EVEN THIS VERY, VERY SIMPLE BRAIN. WHAT DO I MEAN BY THAT? WE CAN POINT TO HUGE REGIONS OF THE BRAIN AND SAY THIS IS THE MEDULLA OBLONGATA OR CEREBELLUM BUT IT'S HARD TO GO TO INDIVIDUAL PARTS OF THE BRAIN AND PICK OUT NUCLEI THAT MIGHT CORRESPOND TO THE AREAS THE HUMAN BRAIN THAT WOULD BE OF INTEREST. SO A COUPLE OF YEARS AGO MY LAB FINISHED UP A PROWHERE WE REGISTERED EXPRESSION PATTERNS FROM 90 REPORTER STRAINS OF TRAN GENIC FISH AND WE WERE ABLE TO CODE REGISTER THEM AT HIGH ACCURACY. WE WERE ABLE TO IMAGE BRAIN OF 90 DIFFERENT LINES OF FISH AND LINE THEM UP SO THE SAME CELL WAS LINED UP IN EACH OF THOSE DIFFERENT BRAIN SCANS. WE THOUGHT THIS MIGHT BE ACTUALLY REALLY GREAT WAY TO TRY AND PRODUCE A MAP OF THE ZEBRAFISH BRAIN. SO OUR IDEA WAS NONE OF THESE 90 PATTERNS UNIQUELY IDENTIFY INDIVIDUAL NUCLEI WITHIN THE ZEBRAFISH BRAIN BUT PERHAPS THERE'S ENOUGH INFORMATION IN THERE THAT SOMEONE REALLY CLEVER COULD GO IN AND WORK OUT TO SUBDIVIDE THAT BRAIN INTO SMALL REGIONS. SO I'M NOT A Pi PERSON AS WARREN DEFINED HOWEVER, I'M LUCKY TO HAVE A GREAT Pi COLLABORATOR, A MATHEMATICIAN/PHYSICIST IN CIT. AND WE PRESENTED HIM WITH THE PROBLEM AND HE SAID THIS LOOKS LIKE A CLUSTERING PROBLEM. HIS IDEA BECAUSE WE HAVE EVERYTHING LINED UP WE CAN TREAT EVERY SECOND QUARTER SEVERAL IN THE BRAIN HAVING A -- VOXEL IN THE BRAIN AS DIFFERENT GENES. THE IDEA IS SIMPLE, YOU NEED TO CLUSTER VOXELS WHICH HAVE A SIMILAR GENETIC IDENTITY. AND SO HE GOT TO WORK ON THAT AND TRIED DIFFERENT CLUSTERING ALGORITHMS. PUTTING IN DIFFERENT WEIGHTING FACTORS BECAUSE OBVIOUS PRISON SEVERALS ADJACENT SHOULD BE IN THE SAME NUCLEUS. THAT WORKED FANTASTICALLY WELL. SO THE RESULT IS HERE ON THE LAST SLIDE. SO THIS ON THE LEFT-HAND SIDE IS NOW THE LARVA ZEBRA FIN BRAIN -- ZEBRAFISH BRAIN AND YOU CAN SEE THE INDIVIDUAL NUCLEI IN THE BRAIN THAT MOST CASES WERE NOT ABLE TO SEE BEFORE. TY EEL GIVE YOU ONE EXAMPLE OF SOMETHING WE DIDN'T KNOW THAT WS THERE BEFORE SO THIS PURPLE BANANA SHAPED REGION IS NUCLEUS OF THE MEDIAN LONGITUDINAL AND ANOTHER GROUP THAT ARDUOUSLY SEGMENTED THAT BY HAND AND OUR LITTLE PURPLE REGION NICELY FALLS WITHIN THE SAME AREA AS THE REGION THEY SEGMENTED. AND IF YOU LOOK AT THIS MAP OF THE BRAIN HERE YOU CAN SEE THAT ONE OF THE COOL THINGS BILATERALLY SYMMETRICAL GIVING THE IDEA THAT THE ALGORITHM IS WORKING BEAUTIFULLY. THE MOST IMPORTANT THING OF COURSE IS THAT WE'RE NOT PICKING OUT REGIONS THAT WE KNEW ARE THERE, WE'RE ALSO PICKING OUT ALL SORTS OF NEW REGIONS SO TAKE A SLICE THROUGH THE BRAIN HERE AND LOOK AT SIDE ON PREVIOUSLY THIS IS WHAT WE HAD IN OUR ATLAS LABELED MEDULLA ON LONGA TA, WHICH IS FANCY WAY OF SAYING HIND BRAIN. NOW WE HAVE THIS DETAILED MAP OF THE REGIONS PRESENT WITHIN THE HIND BRAIN. SO THIS IS GOING TO BE AN UNBELIEVABLY VALUABLE TOOL NOT JUST MY LAB BUT THE MANY OTHER LABS STUDYING THE ZEBRAFISH MODEL THAT ALLOWS US TO TAKE FINDINGS THAT WE GAIN USING THIS VERY SIMPLE MODEL AND TRANSLATE THEM INTO THE HUMAN BRAIN. WE FINALLY HAVE A WAY OF BRINGING INSIGHTS FROM THE SIMPLE FISH TO THE MORE COMPLEX HUMAN. THANK YOU. [APPLAUSE] >> GOOD MORNING, MY NAME IS JONATHAN STREET AND I WILL TALK ABOUT A RECENT PROJECT I'M WORKING ON AIDING THE PATHOLOGISTS WORKING WITH RENAL BIOPSIES. SO WHEN A BIOPSY IS REQUESTED IT IS TO STUDY A STRUCTURE CALLED (INDISCERNIBLE) WHICH LOOKS LIKE THE EXAMPLES ON THE LEFT. THIS IS SURROUNDED BY MAINLY -- SHOWN ON THE RIGHT. FAIRLY BRIEFLY TRAINED OBSERVER TELL THESE ARE DIFFERENT STRUCTURES AND GO INTO THESE IMAGES AND IDENTIFY WHAT TO LOOK AT. BUT A KIDNEY BIOPSY CONTAINS UP TO A HUNDRED OF THESE STRUCTURES SO IT'S QUITE A LABORIOUS TASK TO GO THROUGH AND IDENTIFY EACH ONE INDIVIDUALLY. SO WHAT WE SEE IS THAT THE REPORTING NUMBERS OFTEN UNDERREPRESENT THE TRUE NUMBERS BY UP TO 50%. SO CURRENTLY FOR ROUTINE PRACTICE, THIS ISN'T A MAJOR CONCERN BUT AS WE ATTEMPTED TO DERIVE MORE INFORMATION FROM THESE BIOPSIES, WE WANT THIS NUMBER TO BE AS ACCURATE AS POSSIBLE. SO MY HOPE WAS THAT I COULD APPLY DEEP LEARNING TO STUDY THESE IMAGES IDENTIFY THESE STRUCTURES AND THEN GUIDE THE PATHOLOGIST TO EXAMLY WHERE THEY SHOULD BE FOCUSING. SO I THOUGHT THIS WOULD BE POSSIBLE WHERE INCREASEING UTILIZATION OF -- INCREASING UTILIZATION OF DIGITIZED VERSIONS OF SLIDE SCANNING. THIS ALSO COMES WITH SOME DRAW BACKS IN TERMS OF WE'RE TAKING A SLIDE THAT IS ESSENTIALLY ONE INCH BY THREE INCHES AND WE ARE IMAGING IT WITH SOME MICRON RESOLUTION. SO WE END UP WITH IMAGES THAT ARE APPROXIMATELY 5 BILLION PIXELS. SO WE NEED SPECIAL SOFTWARE TO WORK WITH THESE AND THE FIRST TASK WAS JUST TO DEVELOP A VERY SMALL VERY QUICK PROGRAM TO ALLOW US TO GO INTO THESE IMAGES AND MANUALLY LABEL THEM. THESE LABELS MIGHT BE STORED AS A MASS BUT THEN WE HAVE A SECOND IMAGE THAT IS JUST AS LARGE AS ORIGINAL. SO RATHER THAN TAKING THIS APPROACH I DECIDED TO BORROW ONE FROM GEOGRAPHICAL INFORMATION SYSTEMS WHERE WE CAN STORE MASKS AS GEOMETRYIES THAT TO STORE THE DATA VERY COMPACT TEXT FILE AND ALLOWS OUTS TO DO COMPLEX MANIPULATIONS. SO SHOWN HERE IS ONE EXAMPLE OF A MASK AND AROUND IT IN THE RED SQUARES ARE VARIOUS REPRESENTATIONS WE ARE DRAWING. THEY ALL OVERLAP WITH MINIMUM PERCENTAGE BUT THEY ARE ALL CERTAINLY DIFFERENT. THIS WAS IMPORTANT BECAUSE WE STARTED OFF WITH A FAIRLY MODEST STAGE SET AND GENERALLY SPEAKING WITH DEEP LEARNING IT IS ASSUMED THAT WE NEED A LARGE DATA SET. SO WE GOT AROUND THIS BY AUGMENTING OUR DATA FOR EACH EXAMPLE WE HAD, WE ALTERED SLIGHTLY TO CREATE MORE EXAMPLES AND THEN THE OTHER APPROACH WE TOOK WAS TO USE QUITE A SMALL NEURAL NETWORK FOR DEEP LEARNING FIELD. SO THIS WAS IMAGE DATA SO WE WERE WORKING WITH A CONVOLUTIONAL NETWORK AND IN THIS EXAMPLE WE USED THREE SEPARATE LAYERS AND THEN WE WENT TO A FULLY CONNECTED DENSE NETWORK LAYER. THIS GIVES US AN OUTPUT WHICH WAS ESSENTIALLY A YES, NO RESPONSE, THEN WE COULD APPLY THIS BACK TO OUR IMAGE AND COMPARE IT WITH THE HUMAN ANNOTATION. SO SHOWN HERE WE HAVE THE MACHINE IN RED AND THE HUMAN ANNOTATED IN GREEN AND IN THE LOWER RIGHT HAND CORNER WE HAVE AN EXAMPLE WHERE BOTH HUMAN AND MACHINE CORRECTLY IDENTIFY A GLUE MERLIS. HERE ON THE TOP SENSOR IS AN EXAMPLE WHERE THE MACHINE IDENTIFY IT IS STRUCTURE WE'RE INTERESTED IN BUT THE ANNOTATOR MISSED IT. SO THIS GIVES QUITE A BIT OF CONFIDENCE GOING FORWARD WE SHOULD BE ABLE TO SUPPORT THE PATHOLOGIST IDENTIFYING MORE OF THESE STRUCTURES. IMPROVE OUR ABILITY TO DIVIDE MORE INFORMATION FROM THESE BIOPSIES. THANK YOU VERY MUCH. [APPLAUSE] >> I'M DAVID AND I'M HERE TO TALK ABOUT A LITTLE BROADER MAYBE MORE SPECIFIC PROJECTS A CASE STUDY TALKING THE IDEA OF PAPERS AND FIGURES BEING INTERACTIVE OR DYNAMIC. SO IF YOU'RE SCIENTIFIC PAPERS ARE DISTRIBUTED DIGITALLY, SOME EXCEPTIONS LIKE NATURE AND SCIENCE, STILL PRINT THINGS YOU CAN READ BUT MOST PEOPLE WORK FLOWS YOU SEE PAPER, CLICK ON IT, YOU'RE ON YOUR COMPUTER AND iPAD AND YOU'RE READING IT. AND DESPITE IT BEING ON A COMPUTER, BASICALLY YOU'RE READING -- EVERYTHING IS STACKED YOU HAVE TEXT, IT'S ABOUT THE ONLY CONCESSION -- ABOUT THE ONLY CHANGE FROM BEING ONLINE IS THAT INSTEAD OF HAVING THE TWO COLUMN LAY OUT THERE'S USUALLY ONE COLUMN. BUT YOU'RE STILL -- YOU HAVE THE TEXT AND YOU HAVE THE FIGURES WHICH ARE SOMETIMES EMBEDD WITH THE TEXT OR CLICK A BUNCH BUT THERE'S STATIC. WHICH IS OFTEN DEEPLY FRUSTRATING. SO WE SEE HERE THIS PAPER WHICH I JUST GRABBED FROM A -- THIS IS NATURE REALLY COOL LOOKING FIGURE, IT HAS 644 BLOOD METABOLITE AND YOU HAVE THIS BEAUTIFUL LOOKING FIGURE HERE. AND IT'S FRUSTRATING BECAUSE YOU HAVE ALL THESE DATA POINTS BUT YOU CAN'T ACTUALLY SEE WHAT ANY OF THEM ARE. BECAUSE IT'S A PICTURE. PRESUMABLY DIVE INTO SUPPLEMENTARY DATA AND PULL OUT TABLES FOR THESE BUT IT'S A LOT OF WORK TO DO. SO NOT HERE TO ARGUE THAT NATURE AND PUBLISHERS HAVE TO CREATE A PLATFORM, I'M HERE TO ARGUE THAT IT'S POSSIBLE FOR YOU TO KIND OF CHANGE THIS. SO THIS IS SUPPOSED TO BE A VIDEO. I DON'T KNOW IF I CAN MAKE IT GO. IS THERE A MOUSE SOMEWHERE I CAN CLICK? SO GOOD, YES, A LOT OF TIMES LIKE NEW ORGANIZATION VERSUS TAKEN THE LEAD I FEEL IN MAKING DYNAMIC VISUALIZATION, especially the New York Times. SO INSTEAD OF HAVING ARTICLE TALKING ABOUT WELL, PLACES ARE GETTING WARMER, YOU CAN DEPICT THEIR CITIES. IF YOU'RE TALKING ABOUT HOW SUCCESSFUL TEAM IS DRAFTING PLAYERS, YOU HAVE VISUAL SAYINGS WHERE YOU PICK THINGS AND IN THIS CASE CRIME STATISTICS YOU CAN CLICK AROUND AND PICK THESE TO SEE CRIME STATS SO BASICALLY IN THIS CASE THIS IS GENOMIC DATA SO SOMEONE'S COMPILED RNA SEQ DATA SETS SO CLICK WHAT YOU WANT TO SHOW AND ADDITIONAL FUNCTIONALITY TO UP LOAD BASIC INFORMATION SO HAVING DYNAMIC INFORMATION INCREASES THE IMPACT IN USEFULNESS OF YOUR DATA. KIND OF THE POINT I'M MAKING HERE IS THAT CAN MAKE DYNAMIC VISUALIZATION. DATA IS GETTING BIGGER BASICALLY ACROSS ALL SCIENTIFIC DISCIPLINES. EVERYONE IS COLLECTING DATA AND FIGURES SHOULD BE CODE. AND IF THEY'RE CODE'S NOT A JUMP TO MAKE THEM INTERACTIVE. SO THIS IS A CASE STUDY, THIS IS A PROJECT THAT WAS RECENTLY DONE BY ARCHIVE THEY COMPILED ALL PUBLICLY AVAILABLE HUMAN EYE RNA SEQ DATA SETS AND COMPARED TO THE G TEXT DATA SET SO OVER A THOUSAND SAMPLES IN THIS. PUBLICATION PRESUMABLY SOMEONE READING THE PAPER CAN DOWNLOAD TABLES AND GET INFORMATION BUT THAT WAS SATISFYING SO I TOOK -- SO THESE YOU SEE HERE THESE FIGURES, THIS IS LIVE WEB PAGE -- THIS IS A VIDEO BUT AT THE BOTTOM THERE'S A LINK TO THE WEB PAGE WHICH SHOULD FUNCTION AND YOU CAN BASICALLY SELECT TISSUES YOU WANT, SELECT VIEWS YOU WANT AND BASICALLY GET WHAT YOU WANT OUT OF THE DATA OPPOSED TO WHAT IS SHOWN TO YOU. IN A PUBLICATION. SO THE WAY THIS ALLOWINGS THE READER TO GET DEEPER INSIGHT TO YOUR OWN DATA YOU GENERATED. SO INTERACTIVE TOOL TIPS AND YOU CAN GET A BETTER IDEA WHAT THE DATA LOOKS LIKE. THIS PROJECT WAS DONE -- I DID THE ANALYSIS AND BUILD THE WEB APP, I'M NOT A WEB DEVELOPER, IT'S NOT BECAUSE I'M COLOR IT'S BECAUSE THERE'S GOOD TOOLS OUT THERE, THIS IS BUILT USING R AND SHINY AND YOU CAN GOOGLETOR TUTORIALS ON THESE BUT IT'S NOT HARD TO LAY THESE THINGS OUT. SO THAT WAY YOU CAN HAVE SOME CONTROL OVER HOW YOUR DATA IS PRESENTED VISUALLY. SO THIS CHANGES A MODEL, A MODEL OF HOW WE CAN MAKE OUR DATA MORE USEFUL. THERE'S ALSO SOURCE CODES AVAILABLE FOR IT AS WELL AS ASH CHIEFLING. ALL RIGHT. THANKS. -- AS WELL AS ARCHIVE LINK. THANKS. >> HI, EVERYONE, I'M NICOLAS FIORINI AND I'LL PRESENT PUBMED LABS SANDBOX TOWARD PUBMED 2.0. THERE WERE A LOT OF PEOPLE INVOLVED IN THIS PROJECT BUT I CANNOT PRESENT THEM ALL SO LET'S START WITH INTRODUCTION TO PUBMED. I'M SURE THAT A LOT OF PEOPLE HERE ARE FAMILIAR WHAT PUBMED IS SO THIS IS MORE THAN 27 MILLION PAPERS, MORE THAN 6 MILLION FREE FOLD TEXT AND WE HAVE PREMIUM SEARCHES DAILY. MOST OF THESE SEARCHES ARE SORTED BY DAYTIME BECAUSE THAT'S THE DEFAULT PUBMED AND SMALL FRACTION OF THOSE SEARCHES ARE BY BEST MATCH OR RELEVANCE. YET WE KNOW BEST MATCH IS THE MOST -- MORE USEFUL MODE, MOST USEFUL MODE FOR SEARCHES. AND WE KNOW THAT BECAUSE THERE IS A RATED INCREASE IN RELEVANT SEARCH USAGE IN THE LAST YEAR IN THE LAST EVEN IN THE LAST FEW MONTHS. BECAUSE OF THAT WE WANTED TO IMPROVE THE BEST MATCH OR RELEVANCE MODE AND TO DO THAT WE RELIED ON MA -- MACHINE LEARNING. SO WE RELIED ON MACHINE LEARNING, THAT IS A MODEL FROM AGGRAVATED AND ANONYMOUS PUBMED SEARCH LOGS AND USE THIS MODEL FOR RERANKING RESULTS PROVIDED BY CLASSICAL SEARCH GENERAL, IN THIS CASE SOLAR. THAT PROVIDED PRETTY GOOD RESULTS AND IT'S CURRENTLY ON PUBMED AND WE KNOW THAT IT PROVIDED GOOD RESULTS BECAUSE USERS TENDED TO TAKE HIGHER POSITIONS THAN ON OLDER BEST MATCH SYSTEM. WE WANTED TO GO FURTHER TO IMPROVING USER SATISFACTION AND USAGE AND USER EXPERIENCE. SO WE DEVELOPED PUBMED LABS WHICH IS A SANDBOX. BASICALLY RIGHT NOW CORE FEATURES OF PUBMED AND THAT INCLUDES FOR INSTANCE THE FAST SETS, WE INCLUDED THE SEARCH ASSETS, THE MOST USED CURRENTLY ON PUBMED. ON TOP OF THAT WE WILL BE ADDING NEW FEATURES AND WE WILL TRY THEM OUT AND GET FEEDBACK FROM USERS SO I ENCOURAGE YOU TO TRADE OUT AND -- BECAUSE YOUR ACTIONS IF U WHERE USE NEW FEATURES WE WILL KEEP THEM IN PUBMED 2.0 IF PEOPLE DON'T CARE ABOUT FEATURES WE WILL DROP THEM. WE WILL MOVE FORWARD A NEW VERSION OF PUBMED SO HERE ARE NEW FEATURES. ONE IS DEFAULT ORDER IS BEST MATCH BECAUSE AS I SAID WE KNOW GENERALLY THE BEST MATCH IS MORE USEFUL FOR THE USERS. WE ALSO INTEGRATED SNIPPETS SO PEOPLE CAN GET A BETTER GLANCE AT ARTICLES BEYOND SIMPLE TITLE. WE HAVE HIGHLIGHTING SO PEOPLE UNDERSTAND WHY THESE RESULTS ARE PROPOSED BY FOR A GIVEN PERIOD AND FOR THE ABSTRACT PAGE WE ALSO HAVE BEYOND COMPLETELY REDESIGN INTERFACE, WE HAVE THIS PATIENT NAVIGATION THAT IS STICKY SO THAT YOU CAN ACCESS ANY CONTENT THAT YOU WOULD LIKE FOR THIS PAPER. THIS SIDE BUTTON THAT CONVENIENTLY PROVIDES THIS CITATION INFORMATION, WITHIN VARIOUS PERMITS AND ON TOP OF ALL THAT IT'S COMPLETELY RESPONSIVE SO YOU CAN ACCESS IN PHONES AND TABLETS AND ANY DEVICE. IT'S SMOOTH AND TRANSPARENT SO I ENCOURAGE YOU TO TRY OUT. AND SO THIS IS STILL WORK IN PROGRESS BUT IT WILL BE OFFICIALLY RELEASED IN A FEW WEEKS SO YOU CAN TRADE OUT THIS FULLY WORKING THOUGH IT HASN'T BEEN ANNOUNCED YET. THANK YOU VERY MUCH. [APPLAUSE] >> HI. SO THIS -- I WILL BE INTRODUCING A TOOL TO DO BIOMEDICAL FREE TEXT CLASSIFICATION. THIS TOOL WAS DEVELOPED WITHIN -- WITH A USER WHO WAS A BIOLOGIST BUT NOT REALLY A FUNCTIONAL PROGRAMMER SO YOU WILL FIND THIS TOOL USEFUL IF YOU HAVE CLINICAL BIOLOGICAL TEXT, YOU WANT TO RECOGNIZE AND CLASSIFY THE TERMS IN IT. OFTENTIMES IN TEXT YOU MIGHT HAVE WORDS THAT ARE VERY DIFFERENT, QUITE HARD TO DO STANDARD TEXT STRING PROCESSING ON THEN. FOR INSTANCE ELECTRONIC HEALTH RECORD SAY THE PATIENT HAS HEART ATTACK. OTHER HEALTHY COHORT MIGHT SAY PATIENT HAD MYOCARDIAL INFARCTION. THESE ARE FOR THE SAME CONCEPT, WE WOULD -- SO WE DEVELOPED VERY NICE TOOLS THAT MAP ALL THESE TERMS, VARIANT AND SUCH THINGS INTO A UNIFIED CONCEPT BUT USING THE TOOLS IS A BIT HARD. SO THE PROBLEM FALLS INTO THIS GENERAL PROBLEM OF THE FIELD OF NATIONAL LANGUAGE PROCESSING, WE'RE DOING ADMISSION, IT IS A NICE TOOL DEVELOPED AT NATIONAL LIBRARY OF MEDICINE CALLED METAMAPLITE AND THE TOOL I WILL INTRODUCE EXCEL ON TOP OF METAMAPLITE. HOW DOES THE TOOL LOOK LIKE? YOU GET A SIMPLE ONE FILE, JUST CLICK THROUGH IT, ALL THE COMPILATION IS DONE FOR YOU AND IT INCLUDES ALL THE DIFFERENT LARGE VOCABULARIES THAT ARE NECESSARY TO MAKE IT WORK. IT LAUNCHES A WEB INTERFACE WHICH PROVIDES EXCEL TABLET, POPULATED WITH YOUR FREE TEXT DATA. AND THERE IS -- SIMPLE SINGLE EXCEL FUNCTION WHICH YOU CAN CONFIGURE. AND THE EXCEL FILE ALSO IS SELF-DOCUMENTING AND TELLS YOU HOW TO CONFIGURE THE EXCEL FUNCTION AND IT POPULATES THE CELLS WITH THE TYPES CONCEPT IDENTIFIERS, IT'S EASY TO USE. THE TOOL -- SINCE THIS PROVIDES A WEB API, BECK USE IT -- WE CAN USE IT FOR MAPPING LANGUAGE WITHOUT MAPPING THE MORE COMPLEX API UNDERLYING. IT PROVIDES INTERFACE. THE TOOL IS COMPLETELY FREE. HOW FAR, THE UNDERLYING DATA REQUIRES A UMLS LICENSE WHICH IS ALSO FREE. WE ARE LOOKING AT ALSO RELEASING A TOOL THAT INCLUDES MORE RESTRICTED SUBSET OF UMLS THAT DOES NOT REQUIRE A UMLS LICENSE. IF YOU HAVE A PROBLEM ALONG THESE LINES AND WANT TO MEET FOR A DEMO, TALK TO ME. THANK YOU. [APPLAUSE] >> GOOD MORNING, I'M ROBERT PAWLOSKY, PART OF THE LABORATORY OF METABOLIC CONTROL NIAAA. SO A TOP PRIORITY SET BY THE ASSISTANT SECRETARY OF HEALTH AND HUMAN SERVICES, IN THE NATIONAL PLAN TO ADDRESS ALZHEIMERS DISEASE IS TO PRESENT AND EFFECT TESTIFILY TREAT ALZHEIMER'S DISEASE BY YEAR 2025. WE ARE WELL ON OUR WAY TO ACHIEVING EFFECTIVE TREATMENT IN 2017. IN 2016 THERE WERE OVER 5 MILLION AMERICANS LIVING WITH ALZHEIMER'S INCURRING A HEALTHCARE COST OF $260 BILLION. BY 2050, IT'S ESTIMATED THERE WILL BE 16 MILLION IN AMERICA LIVING WITH ALZHEIMER'S WITH A COST OF $1.1 TRILLION. ALZHEIMER'S IS THE 6TH LEADING CAUSE OF DEATH IN THE UNITED STATES. AND CURRENTLY THERE'S NO CURE AND NO LONG TERM EFFECTIVE TREATMENT. THE BACKGROUND OF THE SLIDE YOU SEE A NEURON. AND NEURONS USUALLY USE GLUCOSE AS AN ENERGY SUBSTRATE. HOWEVER, IF PYRUVATE DEHYDROGENASE IS BLOCKED, THEN GLUCOSE CANNOT GET INTO MITOCHONDRIA. SO MILD CARBOXYLATE SUCH AS KETONES CAN SUPPLY ENERGY RESOURCES FOR THE CELL. HOW DO I GO TO THE NEXT ONE? FOR YEARS OR EVEN DECADES, BEFORE DEVELOPMENT OF CLINICAL SYMPTOMS, ALZHEIMER'S PATIENTS EXHIBIT LOW GLUCOSE UTILIZATION ACROSS VARIOUS REGIONS OF THE BRAIN. BECAUSE OF THE INSULIN INSUFFICIENCY, THIS IS SOMETIMES REFERRED TO AS A TYPE 3 DIABETES. DECREASE INSULIN SENSITIVITY INHIBITS PDH, PYRUVATE DIHYDROGENASE RESULTING IN LOW GLUCOSE METABOLISM, DECREASED ENERGY TO THE CELL, DECREASED BIOSYNTHETIC CAPABILITIES, AND OXIDATION OF MITOCHONDRIA AND POTENTIALLY INCREASE IN BETA AMYLOID AND PHOSPHORYLATED TAU, TWO HALLMARKS OF ALZHEIMER'S. SO THE QUESTION TO US WAS, IS MYOKETOSIS A PRACTICAL THERAPY FOR TREATMENT OF ALZHEIMER'S DISEASE? DURING PROLONGED FAST, MAMMALS BECOME KETOTIC WHERE QUESTION TONE LEVELS INCREASE AS A RESULT OF FATTY ACID OXIDATION IN THE LIVER. IN OUR STUDY, WE CONTROL BLOOD KETONE LEVELS USING A SYNTHETIC KETONE ESTER SYNTHESIZED IN OUR LAB. ALZHEIMER'S ANIMAL MODEL, A TRIPLE TRANSGENIC MOUSE MODEL WAS USED TO STUDY THE EFFECTS OF KETOSIS. ANIMALS WERE PLACED ABOUT MID LIFE ABOUT 8 1/2 MONTHS OLD PLACED ON A NORMAL MOUSE DIET BUT WHERE 15% OF THE CALORIES WERE KETONE ESTER RATHER THAN CARBOHYDRATE. WE CALCULATED CONTROLS WERE MAINTAINED ON THEIR NORMAL DIET. PERIODICALLY ANIMALS WERE SUBJECTED TO A BATTERY OF BEHAVIORAL AND COGNITIVE TESTS. WE USE QUANTITATIVE MASS SPECTROMETRY TO QUANTIFY WIDE RANGE OF METABOLITES FROM THE HIPPOCAMPUS AND FROM THESE CONCENTRATIONS, WE DETERMINE THE CYTOSOLIC AND MITOCHONDRIAL RAY DOCKS POTENTIAL AND DELTA G ATP HYDROLYSIS. HIPPOCAMPAL AND ACETYL ASPARTATE A NEURAL CHEMICAL WITH ANGIOLYTIC PROPERTIES WAS SIGNATURE MITOCHONDRIAL FUNCTION WAS CORRELATED TO SOME BEHAVIORAL OUTCOMES. SO WHAT WE OBSERVED, TO OUR KNOWLEDGE THIS IS THE FIRST TIME A MAJOR PATHOLOGICAL DISEASE INVOLVING A DEFECT IN A KREBS CYCLE CAN BE CORRECTED USING KETONES. WE OBSERVED, IN THIS CHART THE LIGHT GRAY AREAS ARE THE ANIMALS ARE ON THE 15% KETONE ESTER DIET. WE OBSERVED INCREASES ACROSS THE BOARD AND TCA CYCLE METABOLITES. ACETYL COA CITRATE, ISOCITRATE, SUCCINATE AND FUMARATE. WE ALSO OBSERVED ELEVATIONS IN GLYCOLYTIC INTERMEDIATES WHICH YOU CAN SEE IN THIS PORTION OF THIS SLIDE. THERE WAS INCREASED BIOSYNTHETIC ACTIVITY IN THE KETONE ESTER FED ANIMALS AS SEEN AS INCREASE IN ASPARTATE AND INACETYL SPAR AT A TIME, LOWER AMOUNTS OF GABBA WERE ALSO IMPORTANT THEY WILL BE DESCRIBED MORE AT THE POSTER. AS AS FAR AS ENERGY INTERMEDIATES ARE CONCERNED ANIMALS ON THE CONTROLLED DIET HAD HIGHLY OXIDIZED MITOCHONDRIA DETERMINED BY THE NADH RATIO. WHEREAS KETONE ESTER FED GROUP HAD REDUCED MITOCHONDRIA COMPARABLE TO NORMAL TISSUE. THERE WAS ALSO IN A KETONE ESTER FED MILDLY KEETONIC ANIMALS THERE WAS INCREASE IN THE DELTA G OF ATP HYDROLYSIS. FURTHER, THERE WAS ALSO INCREASE IN NADPH. THIS IS IMPORTANT FOR REGENERATING GLUTATHIONE AND DESTROYING TOXIC FREE RADICALS. IN THESE SAME ANIMALS WE OBSERVED THAT THERE WAS IN THE KETONE ESTER FED GROUP COMPARED TO THE CONTROL, THERE WERE DECREASE NUMBERS OF BETA AMYLOID REACTIVE CELLS IN THIS A STAINING TECHNIQUE FOR AMYLOID AND THERE WERE ALSO DECREASED AMOUNTS OF PHOSPHORYLATED TAUO. IN THE KETONE ESTER ANIMALS COMPARED TO THE CONTROL FED. THESE ARE HIPPOCAMPAL SLICES. IN BEHAVIORAL, KETONE MICE ALSO HAD DECREASED ANXIETY AS DEMONSTRATED BY IN THE ELEVATED PLATFORM MAZE SO THIS PART OF THE SLIDE IS THE MICE ARE PUT ON ELEVATED PLATFORM AND THERE'S A CLOSED AREA AND AN OPEN AREA. THE MORE TIME THE ANIMALS SPEND IN THE OPEN AREA, THE LESS ANXIOUS THEY ARE. SO YOU CAN SEE THAT ANIMALS THAT WERE FED THE KETONE ESTER DIET HERE IN THE OPEN ARM PORTION OF THE MAZE WERE -- SPENT SIGNIFICANTLY GREATER AMOUNTS OF TIME IN THE ELEVATED OPEN PORTION OF THE MAZE COMPARED TO THE CONTROLS WHICH WERE SPENDING MORE TIME IN CLOSED AREAS. ALSO THIS WAS HIGHLY CORRELATED WITH THE CONCENTRATION OF HIPPOCAMPAL AND ACETYL ASPARTATE, A KNOWN BIOMARK E FOR ANGIOLYTIC ACTIVITY IN HUMANS. AND TOGETHER ANIMALS THAT WERE ON THE KETONE ESTER DIET SHOWED INCREASED EXPLORATORY BEHAVIOR AS DEMONSTRATED IN THE OPEN FIELD DISTANCE. SO KETONE ESTER ANIMALS DOWN HERE SPENT MORE TIME EXPLORING DISTANCES IN A BOX COMPARED TO THE CONTROL-FED GROUP. RESULTS FROM THIS STUDY SUGGEST MILD KETOSIS SHOULD HAVE POSITIVE BENEFITS FOR TREATMENT IN ALZHEIMERS DISEASE. FURTHER, THIS KETONE ESTER HAS ALREADY BEEN DETERMINED SAFE BY THE FOOD AND DRUG ADMINISTRATION AND HAS BEEN USED IN OTHER HUMAN TRIALS. THANK YOU FOR YOUR ATTENTION. [APPLAUSE] >> I'M CHUNHUA YAN FROM THE CENTER FOR BIOMEDICAL INFORMATICS AND INFORMATION AT NCI. CANCER IS A GROUP OF DISEASE CHARACTERIZED BY UNCULTURE CELL GROUPS. IT IS CAUSED BY GENETIC ALTERATION IN ONCOGENES TUMOR SUPPRESSOR AND DNA REPAIR GENES. IN THE PAST DECADE, NEXT GENERATION TECHNOLOGY HAS GREATLY IMPROVE OUR UNDERSTANDING OF GENETIC LANDSCAPE IN CANCERS. THE NCI AND THE NCI SPONSORED CANCER GENOME ATLAS PROJECT HAS A SEQUENCE OF 10,000 TUMOR SAMPLES IN DIFFERENT CANCERS. AS A MEMBER OF TCGA BREAST CANCER WORKING GROUP TCGA DATA IS AN EXAMPLE. GENOME DATA INCLUDE MUTATIONS AND GENE EXPRESSION COPY NUMBER MICRORNA EXPRESSION AND PROTEOMICS. THE DATA TYPICALLY ANALYZED WITH HIDDEN MAP CLUSTER AND CIRCUS PLOT WHERE TCG IS TREATED ADS INDEPENDENT ENTITY. THAT LEAD US TO DEVELOP OUR -- TO PERFORM INTEGRATED ANALYSIS OF GENOMIC DATA USING A PASSIVE INFORMATION. TODAY I'M GOING TO SHOW YOU TWO FEATURES IN OMICPATH. FIRST NETWORK STRUCTURE AND ENHANCEMENT FROM IMAGING FILES. PATHWAY COME IN DIFFERENT SHAPE AND SIZE. THE PATHWAY HAS APOPTOSIS AND CELL SIGNALING DEPENDING WHICH PATHWAY IS ACTIVATED. AND NCI INITIATIVE CONSTRUCT MOST COMPREHENSIVE PATTERN -- PATHWAY, WHICH IS CONTAINED ABOUT 180 GENES, TYPICALLY PAPER RUSS DESCRIBE THIS AS ONLY 20 TO 30 GENES SO THIS WILL BE VERY DIFFICULT TO COMPARE THIS TO PATHWAYS BUT IT IS EVEN MORE CHALLENGING TO TRY TO MODIFY THEM. OVER THE PAST HAS A FUNCTION -- OMICPATH HAS A FUNCTION TO PASS IMAGE FILE INTO TEXT TABLES. FOR EXAMPLE, IN A TOP TABLE CONTAINING A GENE INFORMATION AND IMAGE, THE BOTTOM CONTAIN A GENE, GENE INFORMATION RELATIONSHIP ACTIVATION OR INHIBITION ALONG WITH THE POSITION INFORMATION. WITH THIS TABLE, OMICPATH IS ABLE TO RECONSTRUCT A PATHWAY AND OVERLAY WITH GENOMIC DATA. FOR EXAMPLE, IN TCGA BREATH CANCER DATA PI 3K IS HIGHLY MUTATED AS A HIGHLIGHTING RED CIRCLE, THE LINE WAS CORRESPOND TO ASSOCIATION OF THE GENE. SO BLUE LINE INDICATE ACTIVATION, RED LINE INDICATE INHIBITION. CLEARLY IT IS NOT SHOWN -- ALL WHOLE PATHWAY IS ACTIVATED SO IT DEPEND ON PATHWAYS ACTIVATED SO IT'S SURVIVAL OR APOPTOSIS. THE SECOND FUNCTION IN THE OMIC PATTERN DISEASE MODEL IDENTIFICATION, YOU PROBABLY FAMILIAR WITH THE REPRESENTATION OF NETWORK. AS MENTIONED EARLIER, NOT EVERY GENE IN THE PATHWAY IS ACTIVATED AS SHOWN IN THE TCGA BREAST CANCER SUBTYPE BASED LIKE ON LUMEN A. SO YOU WANT TO BE ABLE TO IDENTIFY DISEASE MODULE ONLY ACTIVE IN SUBSET. FOR EXAMPLE IN A -- PIK GAMMA AND DELTA IS ACTIVATED BUT NOT IN BASAL TYPE AND IN CONTRAST, EGFR, IS EXPRESSING THE TYPE BUT HIGHLY EXPRESSED IN LIEU MINUTEA. SO THE GENE IN THE DISEASE MODEL CAN BE USED TO BUILD A MODEL TO PREDICT IF IT IS PROGRESSIVE, SURVIVAL AND ADJUST RESPONSE. SO THERE ARE MANY HUGE FUNCTIONS IN OMICPATH SO HOPE YOU FIND THIS USEFUL. THANK YOU. [APPLAUSE] >> GOOD MORNING, I'M KANAKADURGA ADDEPALLI WORKING WITH THE CANCER GENOMICS CLOUD PILOTS. IN SPITE OF RAPID ADVANCES IN THE FIELD OF GENOMICS THERE ARE A LOT OF TECHNICAL DIFFICULTIES WHICH LEAD TO ANALYTICAL BOTTLENECKS FOR RESEARCHERS FOR THE CANCER RESEARCH IN THE CANCER RESEARCH COMMUNITY. IF A RESEARCHER DID NOT HAVE TO THINK ABOUT SCALABLE STORAGE, DATA STORAGE, FASTER DATA TRANSFER, INSTALLING TOOLS, APPLICATIONS, BUILDING WORK FLOWS AND PIPELINES, WE CAN DO A LOT MORE SCIENCE MORE SOONER THAN BEFORE. SO THE IDEA HERE IS FROM THE NCI C BIT FOR THE CANCER GENOMICS CLOUD PILOTS TO DEMOCRATIZE ACCESS TO THE NCI GENERATED GENOMIC DATA AND OTHER DATA RELATED OTHER DATA. AND EFFECTIVELY PROVIDE COMPUTATIONAL RESOURCES AND SUPPORT. THE USERS CAN EASILY ACCESS THESE CLOUD PILOTS AND THEY CAN GET THEIR OWN DATA ALSO. CLOUD PILOTS RIGHT NOW HOST THE MOST COMPREHENSIVE DATA SETS, THE TCGA. ALL THE 11,000 SAMPLES. AND THESE CLOUD PILOTS RIGHT NOW MAKE AVAILABLE MOTHER THAN A COUPLE OF PEDABYTES OF GENOMIC DATA TO IMMEDIATELY TO THE USERS AUTHORIZED USERS. SO THE NCI FUNDED THREE AWARDS TO THREE GROUPS. THE BROAD INSTITUTE WHICH USES GOOGLE CLOUD, IT FOLLOWS THE DATA MODEL APPROACH. WHICH ALSO USES GOOGLE CLOUD AND LEVERAGES THE GOOGLE CLOUD PLATFORM. THEN THERE'S THE SEVEN BRIDGES GENOMICS WHICH USES AMAZON CLOUD. THESE CANCER GENOMICS CLOUD PILOTS CONCEPTUALIZED IN 2014. AND RIGHT NOW WE ARE IN EVALUATION PHASE. SO THESE CANCER GENOMICS CLOUD PILOTS HAVE A LOT OF ADVANTAGES, THEY'RE SCALABLE, THEY HAVE FAST PRE-CONFIGURED TOOLS, WORK FLOWS AND PIPELINES WHICH USERS CAN USE IMMEDIATELY. USERS CAN CREATE ACCOUNTS ON THESE -- ANY OF THESE CLOUD PILOTS EASILY. IF YOU HAVE A GOOGLE GMAIL ACCOUNT YOU CAN USE GOOGLE CLOUD PILOTS JUST BY LINKING YOUR GMAIL ACCOUNT TO IT. LATER ON THESE ACCOUNTS CAN BE LINKED TO YOUR NIH ACCOUNTS ALSO. USERS CAN DO A LOT OF LARGE SCALE ANALYSIS INCLUDING RNA SEQ ANALYSIS, WHOLE EOME AND -- EXOME AND GENOME SEQUENCING, THERE'S LABS YOU CAN USE ON THE FLY, MICRODETECTION PIPELINES FOR EXAMPLE. THESE CANCER GENOMICS PILOTS USE TIERED CREDIT MODEL AND FOR ANY USE RESEARCHER WHO CREATES AN ACCOUNT THEY HAVE AN INITIAL CREDIT -- FREE CREDITS WHICH CAN BE BUMPED UP TO ABOUT NEARLY 1,000 TO $1,300 OF FREE CREDITS WITHIN INCENTIVE OF FEEDBACK. THE CLOUD PILOTS HAVE PRE-CONFIGURED TOOLS AND WORK FLOWS, YOU CAN USE THESE TOOLS TO BUILD YOUR OWN WORK FLOWS AND PIPELINES. THEY HAVE PROGRAMMATIC ACCESS USING THE -- YOU CAN USE -- THEY HAVE APIs AVAILABLE. AND WE ALSO HAVE A POSTER TODAY HERE OUTSIDE SO ANYBODY -- IF ANYBODY IS INTERESTED AND HAS ANY QUESTIONS WE'LL BE HAPPY TO TALK TO THEM. THANK YOU. [APPLAUSE] >> GOOD MORNING, MY NAME IS UMA SHANKAVARAM, I'M FROM BUY INFORMATICS AND RADIATION ONCOLOGY BRANCH. TODAY I WILL TALK ABOUT ONE OF THE PROJECTS WE ARE WORKING ON ON DATA INTEGRATION AND HOW IT CAN BE USED FOR PERSONALIZED MEDICINE IN CANCER. BEFORE I MOVE ON, I WANT TO ACKNOWLEDGE MEMBERS OF MY GROUP AND CANCER UNFORTUNATELY STILL IS THE SECOND LARGEST CAUSE OF DEATH IN THE U.S.. IN 2017 ALONG PROJECTED NUMBER OF CASES ARE GREATER THAN ONE AND A HALF MILLION. OUT OF THEM ONE-THIRD -- THAT TRANSLATES TO 1600 DEATHS. CANCER IS VERY HETEROGENEOUS, CALLING FOR DIFFERENT TYPES OF PERSONALIZED MEDICINE. THESE TYPE OF APPROACHES HELP IDENTIFY PATIENTS BASED ON GENOMIC PROFILES RATHER THAN ON HISTOLOGY. ONE SUCH APPROACH IS CALLED SYNTHETIC LETHALITY. BY DEFINITION SYNTHETIC LETHALITY IS AN ABERRATION WITHIN GENES WHEN PRESENT IN EITHER OF THEM HAS NO AFFECT ON CELL VIABILITY. HOWEVER, COMBINED TOGETHER IT LEADS TO -- THIS PHENOMENA IS WELL ESTABLISHED IN BREAST CANCER PATIENTS WHERE BRCA MUTANT PATIENTS WERE FOUND HIGHLY SENSITIVE TO PARP INHIBITORS. THESE POSITIVE RESULTS LED TO NEW CLINICAL TRIALS AND APPROVAL OF SPEEDY DRUGS IN O VIRGIN PATIENTS WHERE -- OVARIAN CANCERS WHERE PATIENTS ARE IDENTIFIED WITH MUTATIONS. TRADITIONAL METHODS INCLUDE SETTING UP GENETIC SCREENS USING SIR RNA OR CRISPER. ALTERNATELY WE CAN USE BIOINFORMATIC APPROACHES. QUESTION USE THE SECOND METHOD BIG DATA ANALYTICS TO COMBINE MULTIPLE SOURCES OF INFORMATION. FIRST WE WENT TO ABOUT 20 CANCERS FROM TCGA PUT TOGETHER GENE EXPRESSION MUTATION DATA. THEN WE COLLECTED PROTEIN PROTEIN INTERACTIONS AND DRUG TARGET INTERACTIONS. WE SET STATISTICAL TO IDENTIFY PRIORITIZE AND PREDICT SYNTHETIC LETHAL INTERACTIONS. THE DRUG TARGETS AND THEN VALIDATE FOR CLINICAL RELEVANCE. THIS HIGH THROUGH PUT INTEGRATED STUDIES ARE SHOWING POSITIVE RESULTS. JUST BECAUSE SOMETHING IS NOT IN INFORMATIC FOR MOST DOES NOT MEAN IT'S NOT FOR INDIVIDUAL PATIENT. S THIS IS FROM (INDISCERNIBLE) A COMPUTATIONAL BIOLOGIST. WHY DO WE CARE WHAT SHE SAY? SHELLEY SEEMS TO BE SPEAKING FROM PERSONAL EXPERIENCE. STORY GOES IN 2013 SHIRLEY WORKING ON GENE EXPRESSION STUDIES DIAGNOSED WITH ADVANCE STAGE OVARIAN CANCER. HER PROGNOSIS WAS BAD AND SHE WANTED MORE INFORMATION ON HER CANCER. BUT BECAUSE OF OVARIAN CANCER IS SO RARE SHE COULD NOT. SO SHE GREAT SPEECH FROM USC STANDARD DATA ANALYST. SO THE IDEA WAS TO CREATE THE MODEL WHERE THEY CAN USE PUBLIC DATA AND USE HUNDREDS OF OVARIAN CANCER PATIENTS FROM TCGA, IDENTIFY PATTERNS AND SEE WHICH PATTERNS WILL WORK FOR THE TREATMENT AND WHICH WON'T. IN THE MEANWHILE, SHELLEY UNDERWENT CHEMOTHERAPY AND SURGERY AND CHEMOTHERAPY. BUT UNFORTUNATELY HER CANCER CAME BACK HER GENE EXPRESSION STUDIES SHOWED HER PROFILES WERE INDICATING HIGH LEVEL OF IMMUNE RESPONSE GENES. SO USE IT TO TURN ON IMMUNOTHERAPY DRUGS AND THEY DID THAT BUT UNFORTUNATELY STOPPED FOR SOME TIME AND HAS TO UNDERGO SECOND SURGERY AND MORE CHEMOTHERAPY. THE PROGNOSIS WASN'T LOOKING GOOD FOR SOME TIME, SHE TOOK A BREAK FROM TREATMENT AND NEXT APPOINTMENT WITH THE DOCTORS LOOKED AT HER, CIRCULATING TUMOR LEVELS IN HER BLOOD AND STARTED TO SEE THE DECLINE IN THEM. AND THEY CONTINUE TO DECLINE AND SHE WENT INTO RELAPSE. SO SHE STAYED PLEALY RELAPSED TO THIS DAY. THIS KIND OF POSITIVE STORIES MADE US BELIEVE COMBINING PATIENTS SPECIFIC FEATURES DATABASE INFORMATION CAN MAKE A DIFFERENCE AND APPROACHES AND PATIENTS EXPERIENCE HAPPY ENDINGS. THANK YOU. [APPLAUSE] >> HI. I'M STEVE BROOKS, I'M A POST-DOCTORAL FELLOW AT NHLBI, I WORK WITH DR. ACKERMAN IN SICKLE CELL BRANCH, I WANT TO TALK TO Y'ALL ABOUT OUR EFFORTS TO QUANTIFY LOCUST SPECIFIC EXPRESSION ALPHAGLOBIN GENE IN THE VASCULAR EPITHELIUM. ALPHAGLOBIN IS WELL KNOWN AS A SUBUNIT OF HEMOGLOBIN BINDING IN THE TETRAMER WITH BETAGLOBIN AND FOR A LONG TIME BELIEVED PRODUCTION OF ALPHAGLOBIN WAS LIMITED TO RED CELLS. RECENTLY COLLABORATORS AT THE UNIVERSITY OF VIRGINIA DISCOVERED THAT AL HAVEGLOBIN IS PRESENT IN VASCULAR ENDOTHELIAL CELLS. SPECIFICALLY IN THE MYOENDOTHELIAL JUNCTION OF SMALL RESISTANCE ARTERIES ASSOCIATING WITH NITRIC OXIDE SYNTHASE AND IMPACTS NITRIC OXIDE FROM DIFFUSING INTO SMOOTH MUSCLE, A MAJOR PATHWAY CONTROLLING BLOOD PRESSURE AND VASCULAR TONE. OUR MODEL HERE SHOWS THAT WE BELIEVE THAT ALPHAGLOBIN BINDS WITH ENOS IN THE MYENDOTHELIAL JUNCTION AND MOLECULAR OXYGEN QUENCHING THE SIGNAL AND IN THE ABSENCE OF ALPHAGLOBIN ENDOGENOUS NITROUS OXIDE MAY DIFFUSE MORE IN SMOOTH MUSCLE. SO ALPHA GLOW WIN HAS TWO LOCI THAT CONTROL EXPRESSION IN THE GENOME, HBA 1 AND 2. THEY'RE LOCATED ON CHROMOSOME 16 AND IN RED BLOOD CELLS PREVIOUSLY SHOWN THAT THE MAJORITY OF ALPHAGLOBIN IS PRODUCED FROM THE HBA 1 LOCUST. IN FACT ABOUT 70% OF EXPRESSION IS ATTRIBUTED TO HBA 1. WE WANTED TO SEE WHETHER ALPHAGLOBIN IS EXPRESSIONED IN THE VASCULAR ENDOTHELIUM SIMILAR TO IN RED BLOOD CELLS. TO DO THIS WE'RE USING DIGITAL DROPLET PCR. DIGITAL PCR IS A QUANTITATIVE METHOD FOR MEASURING GENE EXPRESSION AND EACH PCR SAMPLE IS DIVIDED TO SMALLER REACTIONS USING A WATER EMULSION TECHNIQUE. ONE 20-LITER REACTION PERFORMING PCR IS DIVIDED TO 20,000 SMALLER REACTIONS, EACH IS IS AMPLIFIED INDIVIDUALLY AND IT'S SHOWN AS POSITIVE OR NEGATIVE RESULT ENABLING A QUANTITATIVE METHOD FOR DETECTING PRESENCE FOR COPIES OF DNA. INSTEAD OF RELATIVE AMPLIFICATION AS YOU MIGHT USE WITH NORMAL QPCR YOU DIRECTLY MEASURE THE AMOUNT OF EACH TRANSCRIPT. THIS TECHNIQUE IS ALSO EXTREMELY SENSITIVE, WHICH IS VERY IMPORTANT FOR OUR PATHWAY. ALPHAGLOBIN ABA 1 AND 2 DIFFER IN RNA TRANSCRIPTS BY ONE BASE PAIR SO WE DESIGNED A 100 BASE PAIR PROBE AND TESTED USING DIGITAL PCR AGAINST SYNTHETIC OLIGONUCLEOTIDES AND CONFIRMED IT IS CAPABLE OF HIGHLY DISTINGUISHING BETWEEN THE TWO GENES. SO WE THEN PROCEEDED TO ISOLATE MICROVESSELS FROM THE BRAIN FROM THE SKELETAL MUSCLE FROM THE KIDNEY AS WELL AS LARGER MUSCLES THE CAROTID AORTA FROM PROFUSED MICE TO ELIMINATE CONTAMINATION. WE WANTED TO SEE IF DIFFERING BETWEEN LARGE VESSELS AND SMALL VESSELS AS WELL AS HOW IT WAS REGULATED IN BLOOD. SO WE MEASURED ALPHA 1 AND 2 AND CONTROLS MEASURED A GENE CALLED AE 1 SPECIFIC TO RED BLOOD CELLS TO ENSURE WE DIDN'T HAVE CONTAMINATION OF RELICK LOW SITES AND ALSO -- RETICK LA SITES AND WE MEASURED NITRIC OXIDE SYNTHASE 3 SIMILAR TO ENDOTHELIUM TO CONFIRM PRESENCE OF ENDOTHELIAL CELLS. WE FOUND IN WHOLE BLOOD ALPHAGLOBIN TRANSCRIPT IS EXPRESSED FROM HBA 1 CONSISTENT WITH THE RESULTS SHOWN IN PREVIOUS STUDIES. WE FOUND THIS WAS RATIO APPROXIMATELY 2.5 TO 1. HOWEVER, IN CONTRAST WE FOUND ABA 2 IS THE PREDOMINANTLY EXPRESSED LOCUST IN THE VASCULAR ENDOTHELIUM TRUE ACROSS VESSELS WITH RATIO OF 1.4 TO 1. THESE FINDINGS WITH OTHER WORK THAT WE'RE DOING IN OUR LABORATORY CURRENTLY STRONGLY SUGGESTS THAT PRODUCTION OF ALPHAGLOBIN IS INDEED DIFFERENTIALLY REGULAR LATED BETWEEN VAS LAR ENDOTHELIUM AND RED BLOOD CELLS. WE HOPE TO TARGET IT IN IMPROVING VASODILATION AND REGULATING BLOOD PRESSURE AND WE MUST BE SURE TO TARGET ENDOTHELIAL ALPHAGLOBIN AND NOT RED BLOOD CELL AND THERE BY IMPACTING HEMOGLOBIN IN ANY WAY. THESE RESULTS OFFER PROMISING INSIGHT IN DIFFERENTIATION BETWEEN THE TWO TISSUES AND OFFER A PATH FORWARD HOW TO EXPLAIN THIS PATHWAY. THANK YOU VERY MUCH. [APPLAUSE] >> WITH THAT WE'LL BRING THIS TO A CLOSE. EVERYONE KNOWS WHERE THE TERRACE IS, IF YOU GO TO THE FRONT OF THE BUILDING, NOT SURE ABOUT THAT, BUT THE ONE THAT -- I GUESS WHAT I THINK OF AS THE MAIN ENTRANCE, T IT'S NOT ON THIS SIDE OF THE BUILDING YOU GO BY THE BOOKSTORE AND RIGHT BEHIND IT IS THE FAES TERRACE. THAT'S WHERE WE CAN ALL JOIN, GET A CHANCE TO TALK WITH THE SPEAKERS AND I WANT TO AGAIN THANK ALL THE SPEAKERS AND LISA FOR ORGANIZING THE PiCo TALKS. THANK YOU VERY MUCH.