>>WELCOME, EVERYONE, TO THE NIGMS JUDITH H. GREENBERG EARLY CAREER INVESTIGATOR LECTURE. I'M JON LORSCH. AS WE BEGIN TODAY'S WEBINAR, PLEASE NOTE THE SESSION IS BEING RECORDED. WE LAUNCHED THE JUDITH H. GREENBERG EARLY CAREER INVESTIGATOR LECTURE IN 2016 IN ORDER TO SHOWCASE NIGMS FUNDED EARLY CAREER INVESTIGATORS AND THE VALUABLE CONTRIBUTIONS THEY'RE MAKING TO PUBLIC HEALTH. THE LECTURE SERIES IS NAMED FOR JUDITH GREENBERG, FORMER DEPUTY DIRECTOR OF NIGMS AND TWO-TIME ACTING DIRECTOR OF THE INSTITUTE WHO HAD DEEP COMMITMENT TO SUPPORTING EARLY CAREER INVESTIGATORS. WE GEAR THESE LECTURES TO STUDENTS TO GIVE THEM THE OPPORTUNITY TO HEAR AND MEET YOUNG SCIENTISTS DOING REMARKABLE THINGS IN BIOMEDICAL RESEARCH. OUR HOPE IS THAT YOU'LL BE INSPIRED TO FOLLOW CAREERS IN THIS AREA. THE LECTURE WILL BEGIN WITH A SPEAKER, DR. CAESAR DE LA FUENTE, GIVING A 30-MINUTE TALK ON HIS RESEARCH AND CAREER PATH FOLLOWED BY A 30-MINUTE QUESTION AND ANSWER SESSION. I ENCOURAGE ALL OF YOU JOINING US ON THE ZOOM TODAY TO ASK QUESTIONS ON DR. DE LA FUENTE'S RESEARCH, CAREER AND SCIENCE BY SUBMITTING THEM TO ME, JON LORSCH, THROUGH THE ZOOM CHAT BOX FUNCTION. SO TODAY'S SPEAKER. AS I MENTIONED, OUR SPEAKER TODAY IS CAESAR DE LA FUENTE. A PRESIDENTIAL ASSISTANT PROFESSOR IN THE DEPARTMENTS OF BIOENGINEERING, PSYCHIATRY AT THE UNIVERSITY OF PENNSYLVANIA. HIS RESEARCH GROUP BELIEVES THAT INNOVATIONS IN ARTIFICIAL INTELLIGENCE MAY HELP TO REPLENISH OUR ARSENAL OF EFFECTIVE DRUGS SUCH AS THOSE USED TO TREAT ANTIBIOTIC RESISTANT BACTERIAL INFECTIONS. SPECIFICALLY, CESAR PIONEERED THE DEVELOPMENT OF THE FIRST ANTIBIOTIC DESIGNED BY A COMPUTER WITH EFFICACY IN ANIMALS, DESIGNED ALGORITHMS FOR ANTIBIOTIC DISCOVERY, REPROGRAMMED VENOMS TO ANTIMICROBIALS, CREATED NOVEL RESISTANCE-PROOF ANTIMICROBIAL TEARS AND DIAGNOSTICS FOR COVID-19 AND OTHER INFECTIONS. HE HAS RECEIVED WIDE RECOGNITION FOR HIS PIONEERING RESEARCH THAT SPANS OVER 100 PUBLICATIONS. NIGMS IS VERY PROUD TO HAVE FUNDED CESAR'S PROGRAM, THE MIRA PROGRAM, SINCE 2020. CESAR, WELCOME AND THANK YOU FOR JOINING US TODAY. >>JON, THANK YOU VERY MUCH FOR THE VERY, VERY KIND INTRODUCTION. I'D LIKE TO START OFF BY EXPRESSING THAT IT'S REALLY A PRIVILEGE TO GIVE THIS LECTURE IN HONOR OF DR. JUDITH GREENBERG, WHO ACTUALLY HAPPENS TO BE HERE IN THE AUDIENCE, SO THANK YOU, DR. GREENBERG, FOR TAKING THE TIME. I REALLY APPRECIATE IT. AND I'M DEFINITELY INSPIRED BY YOUR COMMITMENT THROUGHOUT YOUR CAREER TO ADVANCING WOMEN IN BIOMEDICAL SCIENCE. SO THANK YOU FOR THAT. AND IT'S ALSO HUMBLING TO FOLLOW IN THE FOOTSTEPS OF LUMINARIES IN THEIR RESPECTIVE FIELDS THAT HAVE DELIVERED THIS LECTURE PREVIOUSLY, SO I REALLY FEEL VERY GRATEFUL FOR THIS OPPORTUNITY. I THOUGHT I'D TELL YOU TODAY IS A LITTLE BIT ABOUT OUR EFFORTS AROUND HOW TO USE ARTIFICIAL INTELLIGENCE FOR ANTIBIOTIC DISCOVERY. BUT FIRST BEFORE I BEGIN WITH ALL THE SCIENCE, I WOULD LIKE TO TELL YOU A LITTLE BIT ABOUT MY JOURNEY IN SCIENCE. SO LET'S START FROM THE VERY BEGINNING. THAT'S ME AS A KID, READING. I'VE ALWAYS BEEN VERY CURIOUS ABOUT LEARNING NEW THINGS AND I THINK THAT HAS ALWAYS BEEN A GREAT SORT OF FIRE IN ME TO LEARN. IN SCIENCE, THERE IS NO LACK OF LEARNING AND YOU'RE ALWAYS -- EVEN WHEN EXPERIMENTS DON'T WORK, THAT ALWAYS SERVES AS A LEARNING PROCESS, SO SCIENCE REALLY -- FOR ME, IT MEANS -- IT ENTAILS THE PERFECT FRAMEWORK FOR LEARNING AND BEING CREATIVE AND SO ON. SINCE I WAS A VERY LITTLE KID, I'VE ALWAYS BEEN FASCINATED WITH ENGINEERING AND BIOLOGY. I REMEMBER TRYING TO BUILD FLYING MACHINES SUCH AS THE ONE ON THE RIGHT, FOR EXAMPLE, AND I WOULD CO CONVINCE MYSELF AND MY SIBLINGS TO GET ON THESE FLYING MACHINES AND ONE DAY WE SET IT ON FIRE AND NOTHING HAPPENED. BUT ANYWAYS, ALWAYS VERY CURIOUS ABOUT EXPERIMENTING BOTH WITH PHYSICAL THINGS AND WITH NATURAL THINGS. OTHER THINGS THAT I REMEMBER AS A CHILD, BEING FASCINATED -- I GREW UP BY THE OCEAN, SO BEING FASCINATED BY MARINE ORGANISMS OR JUST PONDERING ABOUT THE ABILITY, THE INCREDIBLE ABILITY OF GE C GECKOS TO REGENERATE THR TAIL. SUCH THINGS WE TAKE FOR GRANTED AS WE BECOME ADULTS BUT THEY'RE REALLY QUITE REMARKABLE, AND THAT JUST SPEAKS TO HOW BIOLOGY AND SCIENCE, YOU KNOW, IT SEEMS LIKE SCIENCE FICTION SOMETIMES, AND I THINK THAT MAKES IT INCREDIBLY FUN FOR ME. SO I DID MY UNDERGRAD IN BIOTECHNOLOGY, SO THAT'S WHERE I LEARNED HOW TO THINK ABOUT SCIENCE A LITTLE BIT AND THAT WAS A VERY FORMATIVE -- VERY FORMATIVE YEARS FOR MYSELF. FROM THEN, I TOOK A LEAP AND I WENT TO UNIVERSITY IN VANCOUVER, CANADA, BEAUTIFUL CITY, AND IN A BEAUTIFUL CAMPUS, AND I HAD THE OPPORTUNITY THERE TO WORK WITH BOB HANCOCK, ONE OF THE PIONEERS IN ANTIBIOTIC DESIGN AND BACTERIAL PATHOGENESIS. AND I HAVE TO SAY A QUICK STORY HERE, I APPLIED TO CANADA BECAUSE IT WAS ONE OF THE FEW PLACES -- I WAS LUCKY TO GO FOR AN INTERVIEW, AND I THINK THE LESSON HERE IS TO HAVE A PLAN B OR C OR D AND I THINK THAT THE NEWER GENERATION IS MUCH MORE PREPARED. I DON'T KNOW WHAT I WOULD HAVE DONE IN MY ENTIRE JOURNEY, REALLY. SO I'M ALSO GOING TO MOVE -- I HAD TO LEARN ENGLISH, AND, YOU KNOW, LEARN A NEW CULTURE AND ALWAYS MOVING FROM ONE PLACE TO ANOTHER AND START FACING SOME OF THESE ADVERSITIES, I THINK HAS HELPED ME A LOT IN SCIENCE. NOT ONLY HOPEFULLY TO BECOME A BETTER SCIENTIST BUT ALSO TO BECOME PERHAPS A BETTER PERSON AND TO ALWAYS TRY TO GROW. THEN AFTER I COMPLETED MY PH.D., I WENT TO MIT WHERE I DID MY POSTDOC, AND THERE I DID MORE -- I WORKED MORE IN SYNTHETIC BIOLOGY AND COMPUTATIONAL BIOLOGY APPROACHES TO ANTIBIOTIC DISCOVERY AND TO MICROBIOLOGY, AND MY TIME THERE OPENED A LOT OF HORIZONS AS TO HOW TO THINK FROM AN INTERDISCIPLINARY PERSPECTIVE ABOUT PROBLEMS AND HOW WE CAN SOLVE THEM BY BRINGING TOGETHER PEOPLE THAT COME FROM COMPLETELY DIFFERENT DISCIPLINES ALL TOGETHER. AND THEN FROM THERE, I WAS VERY FORTUNATE TO BE RECRUITED TO ONE OF THE OLDEST UNIVERSITIES IN THE COUNTRY, UNIVERSITY OF PENNSYLVANIA, THAT WAS FOUNDED BY BENJAMIN FRANKLIN IN 1740, SO IT'S INCREDIBLY OLD, AND IT'S A BEAUTIFUL CAMPUS, IF YOU'VE NEVER BEEN, I RECOMMEND IT. VERY COMPACT CAMPUS WHERE YOU HAVE THE ENGINEERING SCHOOL, THE SCHOOL OF MEDICINE, THE SCHOOL OF ARTS AND SCIENCES ALL TOGETHER, SO WE CAN REACH WITHIN WALKING DISTANCE OF INCREDIBLE COLLEAGUES WHERE YOU CAN SPARC CONVERSATIONS AND INITIATE COLLABORATIONS AND EXECUTE NEW IDEAS. SO THAT'S ABOUT ME BUT I'D BE HAPPY TO DISCUSS ANY OF THIS JOURNEY THROUGHOUT THE Q & A SESSION AT THE END. SO A LITTLE ABOUT THE SCIENCE, WE'RE INTERESTED IN USING ARTIFICIAL INTELLIGENCE FOR ANTIBIOTIC DISCOVERY. DEVELOPING DRUGS IS A VERY DIFFICULT PROCESS. SO AS WE CAN SEE HERE ON THE LEFT, TO DEVELOP A DRUG FROM THE MOMENT THAT YOU DISCOVER IT IN THE LABORATORY, THE FIRST TIME, IN THAT EUREKA MOMENT, TO THE MOMENT WHERE THAT DRUG ACTUALLY HAS AN IMPACT ON PATIENTS IS AN INCREDIBLY LONG AND WINDY ROAD, IT CAN TAKE ON AVERAGE 10 YEARS. SO IT'S A LONG TIME. IT'S ALSO A HIGHLY COSTLY ENDEAVOR. TO DEVELOP A DRUG THERE'S A PREDICTION IT COSTS ABOUT $2.6 BILLION TO DEVELOP A DRUG. THAT'S ACTUALLY MORE THAN THE BUDGET THAT NASA HAS TO TAKE A ROCKET ALL THE WAY TO THE MOON. SO TO DEVELOP A DRUG IS EVEN MORE COMPLICATED ENDEAVOR IN SOME WAYS THAN TAKING A ROCKET ALL THE WAY TO THE MOON, WHICH IS INCREDIBLE TO THINK ABOUT. WE ARE OPTIMISTIC, AND WE THINK THAT THERE ARE A COUPLE OF TRENDS THAT HAVE BEEN GOING ON FOR SEVERAL DECADES THAT CAN HELP ADDRESS SOME OF THESE GAPS THAT WE SEE. SO THE FIRST ONE IS THE EVER-INCREASING COMPUTE POWER. THIS IS FOLLOWING A MOORE'S LAW THAT TELLS US THE NUMBER OF TRANSISTORS WE CAN FIT IN A CHIP DOUBLES EVERY TWO YEARS OR SO, MEANING WE HAVE MORE AND MORE CAPACITY TO PROCESS INFORMATION ON COMPUTERS. AND THE SECOND TREND IS OUR ABILITY TO GENERATE DATA. WE CAN REALLY GENERATE VAST AMOUNTS OF DATA NOWADAYS, AND THAT'S THANKS TO ADVANCES IN AUTOMATION AND IN HIGH-THROUGHPUT SCREENING. SO WE THINK THESE TWO TRENDS CAN HELP US REDUCE THE TIME THAT IT TAKES TO DEVELOP ANTIBIOTICS AND ALSO HELP US REDUCE THE COSTS ASSOCIATE WITH THIS PROCESS. WHY DO WE FOCUS ON ANTIBIOTICS? ANTIBIOTIC RESISTANCE IS A HUGE GLOBAL HEALTH PROBLEM. IT'S CURRENTLY PREDICTED TO LEAD TO THE DEATH OF 10 MILLION PEOPLE PER YEAR BY 2050, SURPASSING EVERY OTHER MAJOR CAUSE OF DEATH IN OUR SOCIETY. THOSE 10 MILLION DEATHS PER YEAR, IF YOU RUN A CRUDE CALCULATION, THEY CORRESPOND TO ABOUT ONE DEATH EVERY THREE SECONDS. SO THIS IS THE FUTURE THAT WE'RE HEADING TOWARDS UNLESS WE COME UP WITH SOLUTIONS, WE COME UP WITH NOVEL TYPES OF ANTIBIOTICS THAT WE CAN USE TO TREAT THESE CURRENTLY UNTREATABLE INFECTIONS. I ALWAYS LIKE TO HIGHLIGHT THAT ANTIBIOTICS ARE NOT ONLY USEFUL WHEN WE HAVE AN INFECTION AND WE TAKE THEM AND WE GET CURED, BUT THEY'RE ACTUALLY ESSENTIAL FOR MODERN MEDICINE AS WE KNOW IT. SO INTERVENTIONS LIKE CHILDBIRTH OR CHEMOTHERAPY TREATMENTS FOR CANCER PATIENTS OR ORGAN TRANSPLANTATIONS OR SURGERIES WOULD NOT BE FEASIBLE WITHOUT EFFECTIVE ANTIBIOTICS. SO IT'S REALLY IMPORTANT TO EMPHASIZE THIS. NATURE, I'M SHOWING A PHOTO OF THE GLOBE HERE BECAUSE NATURE HAS BEEN A GREAT INSPIRATION OF LIKE A SOURCE OF ANTIBIOTICS. FROM THE VERY FIRST ONE, WHICH WAS PENICILLIN, WHICH WAS DISCOVERED BY ALEXANDER FLEMING IN 1928, NATURE HAS REALLY GIVEN US EVERY MAJOR CLASS OF ANTIBIOTICS THAT WE HAVE NOWADAYS IN HOSPITALS. AND SO AGAIN, IT HAS BEEN A GREAT SOURCE OF INSPIRATION AND HAS GIVEN US DIFFERENT CHEMISTRIES THAT WE'VE USED FOR MANY DECADES TO SAVE LIVES, BUT THE PROBLEM THAT WE'RE FACING TODAY IS THAT FOR DECADES, THE SCIENTIFIC COMMUNITY, WE'VE BEEN UNABLE TO FIND TRULY NOVEL CLASSES OF ANTIBIOTICS IN NATURE. SO IN MY LAB, THE WAY WE THINK ABOUT THIS, INSTEAD OF RELYING ON NATURE TO GIVE US ALL THESE LIFE-SAVING MOLECULES, WHY DON'T WE TRY TO TRANSLATE THE CHEMICAL COMPLEXITY OF MOLECULES INTO THE BINARY CODE OF ONES AND ZEROS SO THAT MACHINES CAN TAKE CARE OF THE DISCOVERY PROCESS. AND IN PARTICULAR, WE FOCUS MOSTLY ON SMALL PROTEINS CALLED PEPTIDES IN MY LAB. OF COURSE HOW DO WE ENGINEER PEPTIDES OR PROTEINS? WE FOLLOW VERY BASIC -- THE STRUCTURE OF OUR PROTEIN DETERMINES ITS FUNCTION. SO IF WE CAN CONTROL THE SEQUENCE OF AMINO ACIDS WHICH ARE THE BUILDING BLOCKS OF PEPTIDES AND PROTEINS, WE'LL BE ABLE TO CONTROL THEIR FUNCTION AND THEN CAN YOU TUNE IT. THIS IS A LITTLE LIKE PLAYING LEGO. WE'RE TRYING TO ARRANGE DIFFERENT AMINO ACIDS, BUILDING BLOCKS IN DIFFERENT POSITIONS, TO BUILD SYNTHETIC MOLECULES THAT ARE CAPABLE OF DOING WHAT WE WANT THEM TO DO. SO THIS IS A TRADITIONAL WAY, AND MORE RECENTLY, WE'VE INCORPORATED ASPECTS OF COMPUTATIONAL BIOLOGY AND A.I. TO HELP US ACCELERATE ANTIBIOTIC DISCOVERY. COMPUTERS, THEY CAN HELP US DO A NUMBER OF THINGS, AND I'M GOING TO TELL YOU ABOUT THREE OF THEM. THAT ARE, I THINK, QUITE SIGNIFICANT. THEY CAN HELP US EXPLORE SEQUENCE BASE. AGAIN, SEQUENCE BASE AS I'LL MENTION IS INCREDIBLY VAST FOR ANY MOLECULES BUT INCLUDING ALSO ANY PROTEINS OR ANY PEPTIDES. IT'S ALMOST INFINITE. SO COMPUTERS CAN HELP US EXPLORE THIS. THE SECOND THING IT CAN HELP US DO, MACHINES IS HELP US GENERATE NEW MOLECULES. HERE WE ENTER A LITTLE BIT INTO THE REALM OF CREATIVITY THAT WE TYPICALLY ASSOCIATE WITH THE HUMAN BRAIN, BUT I'LL PROVIDE SOME EARLY EVIDENCE THAT COMPUTERS ALSO -- AND WE CAN TRAIN COMPUTERS TO START TO GET A LITTLE BIT CREATIVE IN TERMS OF THEIR ABILITY TO CREATE NOVEL MOLECULES. LESS BUT NOT LEAST, MACHINES CAN HELP US MINE BIOLOGY TO TRY TO FIND NOVEL DRUGS IN OUR CASE, NOVEL ANTIMICROBIALS. SO I'LL TALK ABOUT EACH OF THESE THREE POINTS. SO SEQUENCE SPACE. HOW DO WE THINK ABOUT SEQUENCE SPACE? SO ILLUSTRATE THIS CONCEPT A LITTLE BIT BETTER, I'M GOING TO SHOW YOU THE SPACE OF CONCEPTS THAT PERHAPS THE AUDIENCE IS MORE FAMILIAR WITH. FOR EXAMPLE, THE NUMBER OF PEOPLE ON EARTH IS 10 TO THE 10. THE NUMBER OF BACTERIA IN HUMAN IS 10 TO THE 13. WE'RE SURROUNDED BY MICROBES, THE MAJORITY OF WHICH DO GOOD THINGS FOR US. IF WE KEEP GOING HIGHER, HIGHER ORDER OF MAGNITUDE, THE NUMBER OF STARS IN THE UNIVERSE, 10 TO THE 31. AND ALMOST UNIMAGINABLE NUMBER. VERY DIFFICULT TO GRASP. AND NOW I WOULD ASK YOU TO CONSIDER A VERY SMALL PROTEIN, A PEPTIDE COMPOSED OF ONLY 25 AMINO ACIDS, IN A LINEAR CHAIN, ONE UP TO THE OTHER, SORT OF LIKE THE COLOR OF PEARLS. I'LL JUST TELL YOU THAT THE COMMUNE TORE YAL SEQUENCE BASE OF THAT VERY SMALL MOLECULE, AGAIN, VERY SMALL, MUCH SMALLER THAN ANY PROTEIN IN OUR BODIES WHICH ARE COMPRISED OF HUNDREDS OF AMINO ACIDS, IT HAS A SEQUENCE BASE THAT IS SUPERIOR TO THE NUMBER OF STARS IN THE UNIVERSE. SO THIS IS INCREDIBLE TO THINK ABOUT, WE NEED COMPUTERS TO START EXPLORING THIS. TO MAKE THINGS MORE COMPLEX AND THEREFORE OF COURSE MORE INTERESTING FROM A SCIENTIFIC PERSPECTIVE, BIOLOGICAL EVOLUTION THROUGHOUT BILLIONS OF YEARS HAS ONLY SAMPLED A TINY FRACTION OF THE ENTIRE SPACE OF POSSIBILITIES OF ALL POTENTIAL MOLECULES, PEPTIDES, PROTEINS. SO EVERYTHING THAT HAS BEEN EXPLORED, WE CAN FIT IT INTO THIS LITTLE PINK OVAL HERE. AND ALL THESE WHITE AREAS REMAIN UNEXPLORED. COMPUTERS CAN HELP US EXPAND THE SEQUENCE SPACE FOR THE WHITE SPACES OR AREAS THAT WE THINK MAY HARBOR MOLECULE SEQUENCES THAT CAN HELP US SOLVE PRESENT-DAY PROBLEMS, FOR EXAMPLE, ANTIBIOTIC RESISTANCE. SO OKAY, CLEARLY I HOPE I CONVINCED YOU THAT WE NEED COMPUTERS TO BE ABLE TO TACKLE SOME OF THESE PROBLEMS AND THE NEXT QUESTION WE ASKED OURSELVES WAS HOW CAN WE TRAIN A COMPUTER TO CREATE DIVERSITY AT THE MOLECULAR LEVEL, TO INNOVATE AT THE MOLECULAR LEVEL? AFTER THINKING ABOUT THIS WITH OUR COLLABORATORS, WE DECIDED THE BEST WAY TO DO THIS WAS TO TRAIN A COMPUTER TO MIMIC THE GREATEST ENGINE WE HAVE FOR INNOVATION AND THAT'S, OF COURSE, NOTHING ELSE THAN EVOLUTION ITSELF. SO WE DECIDED TO TEACH THE COMPUTER HOW TO EXECUTE DARWIN, ON A MACHINE IT BECOMES A THEORY OF ARTIFICIAL SELECTION INSTEAD OF HAVING TO WAIT MILLIONS OF YEARS FOR A MOLECULE TO EVOLVE, WE CAN DO IT ON A COMPUTER IN A TIME SCALE OF DAYS TO WEEKS. THIS IS HOW WE DID IT. WE STARTED WITH AN INITIAL POPULATION OF SMALL PEPTIDES FROM NATURE, AND THEN LIKE I SAID, WE TRAINED THE COMPUTER TO EVOLVE THEM THROUGH MUTATION, SELECTION, RECOMBINATION, WHICH ARE THE ESSENTIAL PROCESSES OF EVOLUTION, AND ITERATIVELY IN A WAY THE COMPUTER IS ABLE TO EVOLVE THE MOLECULES IN REALTIME INSILICO. SO THIS IS WHAT IT LOOKS LIKE, UPON INCREASING NUMBER OF ITERATIONS, COMPUTER EVOLVES THE MOLECULES TOWARD VALUES THAT CORRELATE WITH PREDICTED ANTIBIOTIC ACTIVITY. SO AT LEAST IN PRINCIPLE, THE COMPUTER IS MAKING THE MOLECULES BECOME BETTER AT -- BACTERIA. IN THE PROCESS OF DOING THIS, WHICH IS REALLY INTERESTING, THE COMPUTER IS ALSO CAPABLE OF OF EXPLORING PREVIOUSLY UNEXPLORED SPACES, REMEMBER A FEW SLIDES AGO, THE WHITE SPACES, WHITE AREAS NOT EXPLORED THROUGH EVOLUTIONARY PROCESS, AND YOU CAN START TO INTERROGATE THOSE REGIONS YIELDING MOLECULES DIFFERENT FROM WHAT WE SEE IN BIOLOGY. SO WITH AMINO ACID RATIOS AND COMPOSITIONS THAT ARE DIFFERENT, ATYPICAL, NOT WHAT WE TYPICALLY SEE THAT HAS BEEN GENERATED THROUGHOUT THE EVOLUTIONARY PROCESS. OKAY. SO WHAT I'M SHOWING YOU HERE IS A BUNCH OF DIFFERENT PEPTIDE MOLECULES. YOU CAN SEE THOSE BEAUTIFUL STRUCTURES, BUT EVERYTHING I'M SHOWING YOU HERE JUST GENERATED BY THE MACHINE WITH VERY MINIMAL HUMAN INTERVENTION UP TO THIS POINT. AND HERE, WE REACH A ROADBLOCK VERY COMMON TO ANY PROJECTS INVOLVING COMPUTERS AND THAT IS THAT EVERYTHING THAT WE SEE HERE IS JUST BASED ON COMPUTATIONAL ASSUMPTIONS. IN OTHER WORDS, THE COMPUTER THINKS THAT THESE MOLECULES WILL BE GREAT ANTIBIOTICS, THEY WILL BE GREAT AT KILLING BACTERIA. BUT WE DON'T KNOW THAT FOR SURE, RIGHT? THIS IS VERY MUCH AN EMERGING FIELD SO WE CAN'T REALLY TRUST THE COMPUTER'S ASSUMPTION. WE NEED TO VALIDATE ABSOLUTELY EVERYTHING. SO WE CHEMICALLY SYNTHESIZE THESE MOLECULES AND WE TESTED THEM AGAINST REAL BACTERIA FROM THE HOSPITALS AND FROM THE LABORATORY, TO SEE IF THEY WERE ACTUALLY ABLE TO KILL THEM. SO THROUGH A NUMBER OF SCREENING EFFORTS WE WERE ABLE TO ISOLATE A LEAD PEPTIDE THAT WAS ACTUALLY VERY POTENT AT KILLING PATHOGENS. THEN WE WENT ON TO EE LEWIS DADE THE THREE-DIMENSIONAL STRUCTURE OF THE PEPTIDE WHICH WE SEE, WE CALL THIS GUAVANIN 2. AND OF COURSE WE'RE NOT SEAD WSATISFIED WITH THIS, WE WANTED TO SEE IF THIS COMPUTER-MADE MOLECULE WAS CAPABLE OF REDUCING INFECTIONS IN A REALISTIC CLINICAL MOUSE MODEL. SO WE DEVELOPED THIS SKIN INFECTION MODEL AND WE INFECTED THE MICE AND THEN WE TREATED THEM. WE TREATED THEM WITH TWO OF THE PEPTIDES THE COMPUTER STARTED THE PROCESS WITH, THESE TWO HERE, AND THIS IS THE MACHINE-MADE MODEL, GUA VANIN 2 THAT LED TO BETTER RESOLUTION OF THE INFECTION. THIS IS A CONTROL GROUP OF MICE SO YOU GET A SENSE OF HOW MANY BACTERIA IN THOSE CORE MICE LEFT UNTREATED, YOU CAN SEE THE GUAVANIN 2 MOLECULE IS ACTUALLY CAPABLE OF REDUCING THE INFECTION MUCH MORE EFFECTIVELY THAN EVERY OTHER GROUP. SO JUST TO WRAP UP THIS PART, WE CAN USE COMPUTERS TO ASSIGN NOVEL ANTIBIOTICS NOT ONLY PRETTY TO LOOK AT ON THE SCREEN BUT THEY ACTUALLY KILL, YOU KNOW, PATHOGENIC BACTERIA THAT ARE CLINICALLY RELEVANT IN VITRO AND ALSO IN A MOUSE MODEL. SO THIS IS ANOTHER PROJECT WHERE WE IDENTIFIED A MOTIF, ESSENTIALLY FIVE AMINO ACIDS COLORED IN RED THAT WERE ASSOCIATED WITH TWO FUNCTIONS. DIRECT ANTIMICROBIAL ACTIVITY AND IMMUNOMODULATORY. THE HYPOTHESIS THAT WE REACHED IS, WHY DON'T WE TAKE THIS MOTIF THAT WE PREDICT HAS THESE TWO DIFFERENT FUNCTIONS AND WE'LL TRY TO STITCH IT INTO ANOTHER MOLECULE TO SEE IF IT CAN INVOLVE THAT OTHER MOLECULE WITH TWO FUNCTIONS OF INTEREST. SO IT'S A LITTLE LIKE PLAYING LEGO. SO WE STARTED WITH THIS TEMPLATE. ACTUALLY A TOXIC PEPTIDE, AND WE BASICALLY ENGINEERED IN THE PEPTIDE MOTIF TO ITS END TERMINUS TO CREATE THIS MOLECULE WE CALLED MAST-MO. ESSENTIALLY WE'RE ABLE TO SHOW THROUGH A NUMBER OF ASSAYS IN VIVO AND IN VITRO THAT THAT SYNTHETIC MOLECULE WAS CAPABLE OF DISPLAYING THE DUAL ACTIVITY. SO NOT ONLY THE CONVENTIONAL PARADIGM FOR TREATING INFECTIONS WHERE WITH ANTIBIOTICS YOU TARGET THE PATHOGEN AND YOU CLEAR IT AND THAT RESOLVES THE INFORECAINFECTION BUT ALSO IN AW PARADIGM, WHERE WHAT YOU'RE DOING IS RESOLVING THE INFECTION BY BOOSTING THE HOST'S OWN IMMUNE SYSTEM, IN THIS CASE, THE MOUSE IMMUNE SYSTEM, TO THEN -- YOU BOOST IT IN SUCH A WAY THAT THE MOUSE'S IMMUNE SYSTEM IS ABLE TO CLEAR THE INFECTION. SO WE'RE ABLE TO DEMONSTRATE THIS CONCEPT BOTH IN VITRO IN A SKIN INFECTION MODEL AND ALSO IN A SEPSIS INFECTION MODEL, WHICH IS A HUGE PROBLEM, SEPSIS IS A HUGE PROBLEM IN THE WORLD AS I MENTIONED, HERE ARE SOME OF THE NUMBERS, AND HERE TREATMENT WITH THE SYNTHETIC PEPTIDES WERE CAPABLE OF COMPARING PROTECTION TO MICE AGAINST OTHERWISE LETHAL INFECTIONS OF LABORATORY STRAINS LIKE E. COLI AND STAPH AUREUS BUT ALSO HIGHLY DRUG RESISTANT STRAINS PROBLEMATIC IN THE HOSPITAL, SUCH AS E. COLI KPC POSITIVE AND STAPH AUREUS MRSA -- HIGHLY RESISTANT TO NUMBER OF ANTIBIOTICS. SO THAT WAS ALSO ENCOURAGING AND WE'RE NOW PURSUING FURTHER DEVELOPMENT OF THIS APPROACH FOR TREATING INFECTIONS THROUGH DUAL MECHANISMS. OKAY, SO I TOLD YOU HOW WE CAN USE COMPUTATIONAL TOOLS COUPLED WITH EXPERIMENTS TO DESIGN NOVEL APPROACHES FOR ANTIBIOTIC TREATMENT. OR ANTIBIOTIC DESIGN. NOW I'D LIKE TO TELL YOU HOW WE USE -- BIOLOGICAL INFORMATION, THINGS THAT HAVEN'T REALLY BEEN EXPLORED BEFORE VERY WELL, AND TO TACKLE THIS, WE TAKE INSPIRATION FROM IMAGE AND SPEECH RECOGNITION ALGORITHMS BUT INSTEAD OF RECOGNIZING FACIAL EXPRESSIONS OR SOUNDS, WE WANT TO RECOGNIZE MOLECULAR PATTERNS ASSOCIATED WITH POTENTIAL ANTIBIOTICS. AND WE WORK WITH PEPTIDES, WE LOOK AT AMINO ACID PATTERNS, AND THE ANALOGY THAT I LIKE TO USE HERE IS THAT THESE ALGORITHMS OPERATE A LITTLE LIKE THE SEARCH FUNCTION IN MICROSOFT WORD, WHERE LET'S SAY YOU HAVE A HUGE WORD DOCUMENT LIKE -- WITH HUNDREDS OF PAGES AND YOU WANT TO FIND THE WORD -- ONE PARTICULAR WORD, LET'S SAY NIGMS. YOU GO INTO THE SEARCH FUNCTION, YOU TYPE NIGMS, AND THE ALGORITHM IDENTIFIES OR FINDS THE NIGMS IN WHATEVER INSTANCE IT APPEARS IN THE DOCUMENT. VERY EFFECTIVELY. SOL ALGORITHMS OPERATE IN A SIMILAR FASHION WHERE WE KNOW WHAT WE'RE LOOKING FOR, THE PATTERNS, INSTEAD OF USING LETTERS, WE USE AMINO ACID CODE AND INSTEAD OF SEARCHING OR BROWSING -- WE BROWSE THROUGH GENOMES AND PROTEINS AND WE CAN IDENTIFY POTENTIAL NEW ANTIMICROBIALS. SO TO ILLUSTRATE THIS CONCEPT IN A MORE VISUAL WAY, WE CAN HAVE ESSENTIALLY A PROTEIN IN THREE DIMENSIONS, WE CAN THEN DISPLAY IT IN TWO DIMENSIONS HERE, AND THE ALGORITHM ESSENTIALLY WILL RUN THROUGH THE CODE AND WILL IDENTIFY REGIONS WITHIN THE AMINO ACID CODE THAT ARE PROSPECTIVE ANTIBIOTICS. THAT'S GIVEN IN THE DIFFERENT COLORS HERE, THE DIFFERENT COLORS THAT WE SEE. AGAIN, ONLY A PREDICTION. BUT REALLY A VERY POWERFUL ONE BECAUSE IT ALLOWS YOU TO IDENTIFY WI WITHIN ENTIRE PROTES FRAGMENTS, FOR EXAMPLE THIS FRAGMENT IN YELLOW THAT REPRESENT A POTENTIAL ANTIBIOTIC. WE CAN CONSTRUCT THAT ANTIBIOTIC AND PLAY AROUND WITH IT IN LABORATORY TO SEE IF IT CAN BECOME A POTENTIAL THERAPEUTIC. SO WE'VE BASICALLY UTILIZED THESE METHODS TO EXPLORE THE WHOLE -- TO PERFORM THE FIRST PROTEIN-WIDE EXPLORATION OF THE HUMAN BODY AS A SOURCE OF POTENTIAL ANTIBIOTICS. WE FOUND THOUSANDS OF WHAT WE CALL ENCRYPTED PEPTIDE ANTIBIOTICS AND WHAT I THINK IS REALLY QUITE FASCINATING IS THAT WE DIDN'T ONLY FIND THEM IN THE INNATE IMMUNE SYSTEM, WHICH IS WHERE YOU WOULD IMAGINE YOU WOULD FIND ANTIMICROBIAL MOLECULES. THE IMMUNE SYSTEM HELPS OF FIGHT OFF INVADING ORGANISMS, PATHOGENS. BUT WE WE ALSO FIND THEM ALL THROUGHOUT THE BODY. THIS TAKES US TO OUR CURRENT HYPOTHESIS, WHERE WE THINK THAT PERHAPS IMMUNOLOGICAL RESPONSE IS NOT ONLY RESPONSIBILITY OF THE INNATE IMMUNE SYSTEM, BUT WE HAVE ALL THESE OTHER BODY SYSTEMS WORKING IN COLLABORATION TO PROVIDE -- TO FIGHT OFF INVADING ORGANISMS EITHER DIRECTLY OR INDIRECTLY. SO AGAIN, WE FOUND THOUSANDS OF THESE MOLECULES, AND THEN WE SYNTHESIZE 56 OF THEM USING CHEMICAL METHODS, USING CHEMICAL SYNTHESIS. WE TRY TO LEARN AS MUCH AS POSSIBLE FROM THIS EM. SO I'M GOING TO TELL YOU SOME OF THE LEARNINGS THAT WE TOOK FROM SOME OF THOSE EXPERIMENTS. FOR EXAMPLE, IF YOU TAKEN CRYPTED PEPTIDES THAT ARE ENCODED IN THE SAME BIOGEOGRAPHICAL AREA OF YOUR -- THE SAME AREA OF THE BODY, LET'S SAY THE BLOODSTREAM, AND YOU COMBINE IT IN COCKTAILS, YOU'RE ACTUALLY ABLE TO POTENTIATE THE ACTIVITY TO TARGET PATHOGENS, WHICH IS REALLY INTERESTING. AND WE DEMONSTRATED THIS IN SYNERGISTIC INTERACTION ASSAYS, AND THIS TAKES US TO A VIEW OF SOME OF THESE INCRYPTED MOLECULES WHERE YOU CAN ENVISION HAVING HUNDREDS OF THEM ALL THROUGHOUT YOUR BODY WORKING IN CONJUNCTION TO PROVIDE AN IMMUNOLOGICAL RESPONSE. ANOTHER THING THAT WE WANTED TO SEE IS WHETHER OUR TARGET DEVELOPED RESISTANCE TO THESE INCRYPTED PEPTIDES AND TARGET BACTERIA, SO WE RAN RESISTANCE DEVELOPMENT ASSAYS IN THIS CASE AGAINST GRAM-NEGATIVE HIGHLY DRUG-RESISTANT BACTERIA. AND WE EXPOSED THE BACTERIA OVER TIME, OVER 30 DAYS, CONSISTENTLY, WITH POLYMYXIN B AND THREE ENCRYPTED -- WE FIND IN OUR OWN BODIES. THE BACTERIAL PATHOGEN EVENTUALLY DEVELOPED RESISTANCE TO POLYMYXIN B BUT NOT THE ENCRYPTED PEPTIDES. SO IT SEEMS THE BACTERIA DID NOT READILY DEVELOP PERSISTENCE TO THESE MOLECULES. SO THAT'S ALSO ENCOURAGING. THEN, OF COURSE, WE WENT ON TO SEE WHAT WAS THE INFECTIVE PROPERTY. WE WENT TO ALL THIS INTERESTING TRANSLATING TO SEE IF WE COULD TURN SOME OF THESE MOLECULES INTO ANTIBIOTICS THAT EVENTUALLY THEY HELP PEOPLE. AND WE WERE ABLE TO SEE THAT TREATMENT, MONOTHERAPY TREATMENT WITH ONE PEPTIDE OR ANOTHER PEPTIDE REDUCED THE INFECTION. IN THIS CASE, PSEUDOMONAS INFECTION IN MICE, THIS IS THE UNTREATED CONTROL GROUP, BUT THEN THAT EFFECT WAS EMPHASIZED AND COMBINED IN COCKTAIL, SO WE WERE ABLE TO RECAPITULATE THE DATA WE GOT IN VITRO, ALSO IN VIVO. SO THIS IS COMBINING THESE TWO MOLECULES HERE, WE CAN SEE AN ENHANCED EFFECT AT REDUCING THE INFECTION. AND THE SAME BASICALLY FOR A. BAUMANNIL. BUT THEN WE WENT ON AND WE TESTED THE INFECTIVE EFFICACY IN A PRE-CLINICAL INFECTION MODEL, TYPICALLY THE ONE THAT THE FDA WANTS TO SEE FOR -- ANTIBIOTICS. ONCE AGAIN WE SEE THE TREATED CONTROL GROUP, WE HAVE AROUND 10 TO THE 9 BACTERIA IN THOSE MICE, AND THEN WE HAVE MONOTHERAPY WITH ONE PEPTIDE, MONOTHERAPY WITH ANOTHER PEPTIDE AND THEN COMBINATION THERAPY. SO WE SEE THE POWER, THE FORCE OF COMBINING COCKTAILS AND HOW THAT ACTUALLY ENHANCES THE ANTIMICROBIAL ACTIVITY IN VIVO VERY SUBSTANTIALLY. THIS IS JUST TO SHOW THAT SOME OF THESE ENCRYPTED PEPTIDES ACTUALLY PRODUCE -- IN THE BODY. WE DID IT THROUGH A LITERATURE SEARCH AND WE HAVE FOUR EXAMPLES HERE, THE CUB DOMAIN 3, FIBROBLAST GROWTH FACTOR 5, NATRIURET IC PEPTIDE AND VON WILLEBRAND FACTOR, FROM DIFFERENT LEVELS, LOW TO MEDIUM TO HIGH, IN DIFFERENT AREAS OF THE BODY. THIS IS SUPPORT THAT THESE MOLECULES ARE PRODUCED NATURALLY IN OUR BODIES, FRAGMENTS OF LARGER PROTEINS, AND THEY ARE CONFERRING IMMUNOLOGICAL EFFECT. SO I'M GOING TO SHIFT GEARS A LITTLE BIT. SO ONE -- THIS IS A FUN COLLABORATION WITH A COLLEAGUE OF MINE IN BARCELONA, AND THE CONCEPT HERE IS THAT ANTIBIOTICS ARE VERY PASSIVE MOLECULES. WE INTRODUCE THEM INTO, FOR EXAMPLE, THE BLOODSTREAM AND THEY GO ALONG FOR THE RIDE THROUGH THE BLOODSTREAM. THEY DON'T HAVE ANY SORT OF DIRECTION. THEY DON'T KNOW WHERE THE TARGET SITE IS, SO WE DECIDED TO SEE IF WE COULD CONFER ANTIBIOTICS WITH SOME SORT OF DIRECTIONALITY. AND WHAT WE DID IS WE USED NANOMACHINES AND WE LOADED THEM WITH ANTIBIOTIC AND THE NANOMACHINES CAN SELF-PROPEL THROUGH A DISTANCE, AND THE ANALOGY HERE WOULD BE LIKE IF YOU HAD A TRACK, RIGHT? YOU'D LOAD IT WITH A BUNCH OF ANTIBIOTIC AND THEN THE TRACK CAN MOVE AND IT CAN TRANSPORT ANTIBIOTIC. AND THEN AS IT TRANSPORTS ANTIBIOTIC, THE ANTIBIOTIC IS CAPABLE OF RESOLVING THE INFECTION, BUT IN A DIRECTIVE WAY, IN AN AUTONOMOUS WAY. SO WHAT YOU'LL SEE HERE IN THE DIFFERENT QUADRANTS, YOU'LL SEE MOVEMENT OF NANOMACHINES THAT ARE BEING SELF-PROPELLED AND TRANSPORTING ANTIBIOTIC AND CLEARING AN INFECTION, AND IN THIS QUADRANT HERE, YOU WON'T SEE MATCH BECAUSE THE SUBSTRATE THAT DRIVES THE INTERACTION IS NOT PRESENT SO ALL YOU'LL SEE IS A BIT OF -- MOTION IN THIS QUADRANT. SO HERE'S A LITTLE VIDEO. YOU CAN SEE IN MOST CASES YOU HAVE DIRECTIONALITY AND THE NANO MACHINES ARE MOVING ANTIBIOTIC, I THINK THIS OPENS AVENUES FOR THINKING ABOUT IN DESIGNING ANTIBIOTICS THAT CAN REACH THE TARGET SITE IN AN AUTONOMOUS FASHION AND THAT WOULD AVOID A LOT OF THE SIDE EFFECTS WE HAVE WITH ANTIBIOTICS THAT GO ALL THROUGHOUT OUR BODIES. THEN JUST TO WRAP UP A LITTLE BIT, SO WE'RE REALLY FASCINATED BY COMPUTERS AND THEIR ABILITY TO DESIGN AND DISCOVER NOVEL DRUGS, FOCUSED ON ANTIBIOTICS, AND NOW WHAT WE WOULD LIKE TO DO IS WE'D LIKE TO STREAMLINE THIS SO THAT ROBOTS CAN MAKE THE MOLECULES THE COMPUTER TELLS THEM TO MAKE. AND THEN STRE STREAMLINE THIS FURTHER, CONNECT THEM TO A SCREENING FACILITY THAT ALLOWS US TO SCREEN THOSE MOLECULES FOR ANTIBIOTIC ACTIVITY OR OTHER TYPES OF ACTIVITY THAT YOU MIGHT BE INTERESTED IN, WHETHER CANCER OR OTHERS. WE'RE BUILDING MACHINE LEARNING MODELS TO TIE BACK THE FUNCTIONAL INFORMATION FROM THOSE BIOASSAYS BACK TO THE COMPUTER SO WE CAN HAVE A SELF-LEARNING PLATFORM THAT ALLOWS US TO DISCOVER A NOVEL -- NOVEL ANTIBIOTICS REALLY IN A VERY RAPID MANNER. THIS IS JUST TO ILLUSTRATE HOW YOUNG THIS FIELD IS. AT THE INTERSECTION OF A.I. AND ANTIBIOTIC DISCOVERY. THIS IS A RETROSPECTIVE STUDY WE RUN USING THE QUERIES -- CONTINUING THE QUERIES OF ANTIBIOTICS AND A.I., CANCER THERAPEUTICS A.I. AND DRUGS A.I. YOU CAN SEE WE HAD PRACTICALLY NO PUBLICATIONS UNTIL 2018 SO REALLY WE WERE IN THE MIDST OF A YOUNG AND EMERGING FIELD. I ALWAYS LIKE TO -- INTERDISCIPLINARY RESEARCH COMING FROM BIOLOGY, MICROBIOLOGY, PHYSICS, COMPUTER SCIENCE, DOESN'T REALLY MATTER, CHEMISTRY, THAT IT WOULD BE WONDERFUL TO HAVE YOU JOIN THESE EFFORTS. AGAIN -- AFFECTS EVERY CORNER OF THE WORLD AND THEY'RE REALLY MUCH NEEDED SO REALLY TRIED TO USE YOUR TALENT AND YOUR INGENUITY AND YOUR PASSION TO TRY TO MOVE THIS FIELD FORWARD WOULD BE REALLY FANTASTIC. SO IF ANYBODY WANTS TO GET IN TOUCH, PLEASE DO SO. I'D LIKE TO ALSO MENTION THEY ASKED ME TO ADD THIS TO MY TALKS AND THIS IS A BOOK THAT WE RECENTLY PUBLISHED ON MACHINE LEARNING FOR DRUG DISCOVERY. I THINK IT SERVES AS A GOOD PRIMER FOR BEGINNERS TO TRY TO GET INTO THIS FIELD SO IF ANYBODY IS INTERESTED, I'M ABLE TO GET A FREE COPY. I'D LIKE TO END HERE. I'D LIKE TO THANK THE LAB MEMBERS. IT'S REALLY A PRIVILEGE TO WORK WITH THEM. HIGHLY INTERDISCIPLINARY PEOPLE, VERY SMART. A LOT OF THE THINGS I'VE SHOWN TODAY WAS DONE BY PEOPLE IN MY LAB AND MY COLLABORATORS, SO REALLY OUTSTANDING. OUR FUNDERS FOR ALLOWING US TO DO THE WORK WE DO AND PRIMARILY I'D LIKE TO THANK NIGMS FOR THEIR SUPPORT, REALLY INSTRUMENTAL AT A CRUCIAL POINT IN MY CAREER, REALLY MADE ME BELIEVE THAT I COULD ATTRACT FUNDS TO FUND SOME OF OUR IDEAS. SO THAT WAS A REALLY KEY MOMENT. HERE JUST LINKS TO OUR WEBSITE, EMAIL ADDRESS AND TWITTER HANDLE IN CASE ANYBODY WANTS TO GET IN TOUCH THROUGH ANY OF THOSE MEANS, I'M ALWAYS HAPPY TO CHAT ABOUT SCIENCE OR CAREER ADVICE OR ANYTHING LIKE THAT SO ONCE AGAIN IT'S REALLY AN HONOR TO HAVE DELIVER THIS LECTURE AND THANK YOU FOR YOUR ATTENTION AND I'D LOVE TO TAKE ANY QUESTIONS AND HAVE A DISCUSSION WITH THE AUDIENCE. THANK YOU. >>THANK YOU SO MUCH, CESAR. THAT WAS REALLY TERRIFIC. WE HAVE SOME QUESTIONS. WE'LL START WITH A VERY GENERAL ONE. HOW DID YOU GET THE IDEA TO WORK ON THIS PARTICULAR VERY FASCINATING AREA OF RESEARCH? >>THAT'S A GREAT QUESTION. I THINK MY PH.D., I SAW IT BEING VERY MUCH TRIAL AND ERROR AND I THOUGHT THEY HAD TO BE A BETTER SORT OF ENGINEERING SOLUTION TO IT THEN WITH THE ADVANCES IN COMPUTE POWER AND COMPUTER GENERATION I THOUGHT COMPUTATIONAL TOOLS WOULD BE THE BEST TOOL FOR THE PROBLEM. I'M INCREDIBLY PASSIONATE AND THAT'S WHAT TOOK ME TO SORT OF MORE TRIAL AND ERROR RESEARCH AT THE TIME WHERE I HAD TO ENGINEER A PEPTIDE AND TRY IT OUT AND SO ON, NOW WE CAN DO IT ON THE COMPUTER IN A MUCH FASTER SCALE AND IT REALLY ACCELERATES EVERYTHING, MANY, MANY [INAUDIBLE] >>GREAT. LET'S SEE, WE GOT A NUMBER OF MORE QUESTIONS HERE. DO YOU HAVE ANY ADVICE FOR UNDERGRADUATES WHO ARE INTERESTED IN DOING RESEARCH, AND CERTAINLY YOU SUGGESTED PEOPLE COULD COME WORK WITH YOU, BUT IN GENERAL, WHAT KIND OF ADVICE WOULD YOU HAVE FOR UNDERGRADUATES WHO WANT TO GET INTO THIS KIND OF AREA? >>I WOULD SAY FOR UNDERGRADS, YOU'RE AT A STAGE IN YOUR LIFE, YOUR CAREER, THAT I THINK IT'S VERY IMPORTANT TO EXPLORE DIFFERENT OPTIONS, TO EXPLORE REALLY WHAT CALLS YOU, WHAT YOU REALLY LIKE DOING DAY IN AND DAY OUT, WHAT YOU'RE TRULY PASSIONATE ABOUT. IF YOU'RE INTERESTED IN RESEARCH, EXPLORE DIFFERENT RESEARCH AREAS, GO TO DIFFERENT LABS FOR THE SUMMER AND THEN TRY TO FIND SOMETHING -- YOU MIGHT BE LUCKY, YOU GO TO THE FIRST ONE AND YOU LOVE IT AND THAT'S WHAT YOU DO THE REST OF YOUR LIFE. BUT THAT MIGHT NOT BE THE CASE AND THAT'S OKAY TOO. YOU MIGHT NEED TO EXPLORE DIFFERENT THINGS. COMPUTATIONAL RESEARCH IS VERY DIFFERENT FROM EXPERIMENTAL RESEARCH. SOME PEOPLE LIKE ONE AND NOT THE OTHER, SOME PEOPLE LIKE BOTH, SO YOU HAVE TO FIND YOUR OWN CALLING WITHIN RESEARCH. RESEARCH IS INCREDIBLY WIDE ENDEAVOR SO THERE'S A PLACE FOR EVERYBODY AND I THINK YOU JUST HAVE TO EXPLORE IN ORDER TO FIND YOUR PLACE IN IT. >>GREAT. WE'VE GOT A NUMBER OF QUESTIONS SORT OF RELATED TO EACH OTHER BUT THE FIRST IS WHAT DO YOU KNOW ABOUT THE MECHANISM OF ACTION OF THESE PEPTIDES? ARE THEY ALL THE SAME, ARE THEY DOING DIFFERENT THINGS? DO YOU HAVE SPECIFIC EVIDENCE? >>FANTASTIC QUESTION. SO A LOT OF THEM, THEY TARGET THE MEMBRANE. SO ONE THING I DIDN'T GET INTO IN THE TALK BUT THE EVOLUTIONARY ALGORITHM WE USE IS DRIVEN BY A FITNESS FUNCTION THAT ESSENTIALLY SELECTS FOR MINIMAL PEPTIDE STRUCTURES THAT WILL TATARGET BACTERIAL MEMBRANE. WE USE DESCRIPTORS THAT YOU NEED TO HAVE SOME POSITIVE CHARGES IN THE PEPTIDE AND HYDROPHOBIC AMINO ACIDS. THE POSITIVE CHARGES HAD, THEY ALLOW THE PEPTIDE TO ELECTROSTATICALLY CHARGE WITH BACTERIAL MEMBRANE, THAT ALLOWS TO APPROACH THE MEMBRANE AND IT ALLOWS TO TRANS LOCATE INTO THE MEMBRANE CREATING A PORE THAT TYPICALLY LEADS TO CELL DEATH. SO THOSE ARE SOME OF THE CONCEPTS THAT WE'RE USING TO BUILD MOLECULES THAT WILL TARGET AND KILL BACTERIA. >>THAT SEGUES TO THE SECOND QUESTION WE'VE GOTTEN A NUMBER OF EXAMPLES OF, GIVEN MANY OF THEM SEEM TO WORK BY MAKING A WHOLE IN THE MEMBRANE ESSENTIALLY, WHAT ABOUT TOXICITY? SOME OF THESE EXPERIMENTS WERE IN MICE, SOME WERE IN CULTURE, I GUESS, BUT WHAT DO YOU KNOW ABOUT TOXICITY OF THE VARIOUS COMPOUNDS AND CAN YOU USE A.I. TO PREDICT IT I GUESS AS A FOLLOW-UP TO THAT >>FASCINATING QUESTION. TOXICITY, OF COURSE BEFORE WE TRANSITION ANY OF THE MOLECULES TO MICE, WE ALWAYS TEST FOR TOXICITY AGAINST MAMMALIAN CELL LINES OR MOUSE CELL LINES. ANYTHING TO MICE IS NON-TOXIC BUT SOMETIMES WE COME ACROSS POP TIDES THAT ARE TOXIC SO WE RULE THEM OUT. WE HAVE MACHINE LEARNING MODELS THAT ALLOW US TO PREDICT TOXICITY AS WELL AND THE HOPE IN THE FUTURE IS THAT AS PART OF OUR ALGORITHM, WE ARE NOT ONLY ABLE TO OPTIMIZE TO MAKE THE MOLECULES BETTER AT THAT BUT HOPEFULLY WE CAN ALSO OPTIMIZE TO COUP TER SELECT FOR CYTOTOXICITY AND THINGS LIKE THAT THAT WE DON'T WANT. THE DREAM THAT WE HAVE IS TO CREATE ANTIBIOTICS SO TO HAVE SOMETHING THAT IS ACTIVE, NON-TOXIC, SUFFICIENTLY STABLE, THAT HAS GOOD PHARMACOKINETIC, PHARMACODYNAMIC PROPERTIES AND SO ON. SO THE OVERARCHING GOAL IS TO BUILD A MOLECULE OR SUBSET OF MOLECULES THAT INCORPORATE ALL THOSE ATTRIBUTES. >>GREAT. THAT SEGUES INTO A QUESTION BY IAN MORGAN, HOW DO YOU TRAIN THE MODELS TO DETERMINE WHICH COMPOUNDS WOULD MAKE THE BEST ANTIBIOTICS? I THINK YOU KIND OF SHOWED A FEW PARAMETERS, BUT HOW DID YOU DETERMINE WHICH ONES TO USE AND HOW TO USE THOSE TO TRAIN THE MODELS? >>GREAT QUESTION. FOR THE GENETIC ALGORITHM, WE USED ISOMERS -- LOOKING AT HYDROPHOBIC AMINO ACIDS AND BASED ON -- SCALE, SO AGAIN THIS IS JUST TO BUILD A MINIMAL STRUCTURE -- THE BACTERIAL MEMBRANE, BUT SINCE THEN, WE'VE CONTINUED THAT WORK AND NOW WE'RE USING MANY DIMENSIONS OF PHYSIOCHEMICAL DESCRIPTORS AT A TIME, SO UP TO 500 OR MORE. BASICALLY AS MUCH INFORMATION AS POSSIBLE TO DESCRIBE AN AMINO ACID WE CAN SEQUENCE. >>GREAT. IN DEVELOPING YOUR MODELS, YOUR A.I. MODELS FOR ANTIBIOTICS, HAVE YOU OR THE MODELS FOUND NOVEL TRENDS IN PROTEIN DESIGN THAT YOU HAVEN'T SEEN BEFORE OR WERE SURPRISED TO FIND OUT WAS MORE IMPORTANT THAN YOU PREVIOUSLY THOUGHT? >>INTERESTING QUESTION. I THINK A LOT OF THE PEPTIDES THEY TEND TO BE HELICAL WHEN IN CONTACT WITH THE MEMBRANE BUT WE HAVE FOUND SOME BENEFITS IN THERE AND DIFFERENT SECONDARY STRUG STRUCTURES. NOT EXACTLY RELATED TO THIS BUT ONE THING I FOUND QUITE INTERESTING IN EVOLUTIONARY ALGORITHM, IF WE ACTUALLY LET THE ALGORITHM GO FOR TOO LONG AND REACH A PLATEAU IN TERMS OF ITS OPTIMAL SOLUTION, IT STARTED BASICALLY SPITTING OUT SEQUENCES THAT WERE VERY SIMILAR TO WHAT NATURE HAS CREATED. SO DIVERSITY AT THE MICRO LEVEL WE HAVE TO STOP THE ALGORITHM BEFORE IT REACHES A PLATEAU AND THAT'S WHERE IT STARTED GENERATING DIVERSITY. I STILL DON'T KNOW WHAT TO MAKE OF THOSE RESULTS BUT I THINK IT'S AN INTERESTING -- IT'S INTERESTING CONCEPTUALLY TO THINK ABOUT THAT. >>INTERESTING. I'M SURE YOU'RE AWARE OF ALPHA FOLD, THIS IF YOU APPROACH TO PREDICTING PROTEIN STRUCTURE USING ARTIFICIAL INTELLIGENCE MACHINE LEARNING. HAS THAT APPROACH INFLUENCED YOUR WORK AT ALL? >>YEAH, I THINK IT'S A FANTASTIC TOOL FOR PROTEIN STRUCTURE PREDICTION. FOR PEPTIDES IT DOESN'T WORK THAT WELL. WE'VE TRIED A COUPLE TIMES BUT IT'S JUST BECAUSE THE ALGORITHMS WERE NOT TRAINED ON PEPTIDE DATA, SO THEY CAN REALLY -- THEY CAN'T REALLY ELUCIDATE THE STRUCTURE OF SOMETHING THAT THEY'VE NEVER SEEN. YOU KNOW, A.I. IS NOT QUITE YET AT THAT LEVEL, SO WE'VE TRIED TO USE IT FOR A COUPLE THINGS, WE HAVEN'T -- NOT VERY SUCCESSFULLY, BUT I DO THINK IT'S AN INCREDIBLE TOOL FOR SCIENTISTS ALL AROUND THE WORLD WORKING ON PROTEINS AND I'M PRETTY SURE SOON IT WILL BE APPLIED TO SMALLER PEPTIDES AS WELL AND IT'S Q.US A MATTER OF TIME. >>A NUMBER OF PEOPLE WANT TO KNOW WHAT THE NEXT STEPS ARE FOR THIS REALLY PROMISING PEPTIDES YOU HAVE IN TERMS OF REALLY GETTING THEM INTO THE CLINICS. >>THAT'S EVENTUALLY THE DREAM THAT WE HAVE. FOR A LOT OF THEM, WE PROBABLY NEED TO DO SYSTEMATIC TOXICITY STUDIES JUST TO MAKE SURE AND SYSTEMATIC AND PROBABLY PKPV, PHARMACOKINETIC PHARMACODYNAMICS, HOW THE PEPTIDES ARE DISTRIBUTED THROUGHOUT THE BODY. WE HAVEN'T DONE THOSE STUDIES IN DETAIL. SO THOSE ARE PROBABLY NEXT STEPS. THEN AFTER THAT, OF COURSE, YOU HAVE IND ENABLING STUDIES, NEW DRUG STUDIES, AND THEN AFTER THAT IF THAT'S SUCCESSFUL, YOU GO TO PHASE 1 CLINICAL TRIALS, PHASE 2, PHASE 3, WHERE ESSENTIALLY WHAT YOU LOOK FOR IS SAFETY AND EFFICACY IN HUMANS. BUT WE'RE STILL NOT THERE. >>IS THERE ENOUGH DATA ON PK-PD AND PEPTIDES FOR YOU TO USE AN ARTIFICIAL INTELLIGENCE APPROACH TO PREDICT WHICH ONES WOULD HAVE THE BEST PROPERTIES? >>UNFORTUNATELY NOT. THAT'S SOMETHING I WOULD LOVE TO COMBINE, COMPILE, PUT TOGETHER A DATASET THAT HAS SUFFICIENT PKPB DATA, BUT THERE ARE VERY FEW STUDIES, USUALLY USING RADIOACTIVE APPROACHES AND NOT A LOT OF PEOPLE DO IT AND THERE'S A SCARCITY OF DATA. SO THAT'S THE ROADBLOCK THAT WE'RE REACHING IN A LOT OF THESE THINGS, IS THAT YOU ACTUALLY NEED TO HAVE QUITE A LOT OF DATA. WE GENERATE A LOT IN HOUSE AND SO WE FOCUS ON A COUPLE THINGS AND THEN WE GENERATE THE DATASETS WE THEN USE TO DEVELOP OUR MODELS. BUT RELYING ON PUBLIC DATABASES SOMETIMES NOT -- THEY'RE NOT THE MOST RELIABLE BECAUSE PEOPLE DO THINGS DIFFERENTLY USING DIFFERENT MEDIA, DIFFERENT METHODS AND SO -- BUT WITH PKPB, WE HAVE THAT HUGE ROADBLOCK WHERE WE DON'T HAVE ENOUGH DATA. IT'S ACTUALLY VERY EXPENSIVE AND VERY DIFFICULT TO GENERATE PKP -- DATA. >>THERE'S A QUESTION ABOUT GRAM POSITIVE DATA. YOU SAW MOSTLY GRAM-NEGATIVE. DO YOU HAVE DATA FOR GRAM POSITIVE AS WELL? >>WE HAVE DATA FOR GRAM POSITIVE. ANOTHER BIG PROJECT WE HAVE IN THE LAB IS THAT WE'RE TRYING TO ALSO DESIGN MOLECULES THAT SPECIFICALLY TARGET GRAM-NEGATIVE AND NOT GRAM POSITIVE AND VICE VERSA AND EVEN TO MORE GRANULAR LEVEL LIKE SPECIFICITY LEVEL, GINO SPECIFICITY AND SO ON. SO THIS IS A BIG PROJECT WE'RE CURRENTLY CARRYING OUT. >>THAT BRINGS UP ANOTHER SET OF QUESTIONS ABOUT WHAT HAPPENS TO THE GUT MICROBIOME OR WHAT WOULD YOU THINK WOULD HAPPEN WHEN YOU USE THESE PEPTIDES? >>THAT'S A GREAT QUESTION. OF COURSE YOU DON'T WANT TO KILL THE GUT MICROBIOME, RIGHT? WHEN YOU'RE TAKING AN ANTIBIOTIC. THAT'S ONE OF THE UNINTENDED CONSEQUENCES THAT A LOT OF CONVENTIONAL ANTIBIOTICS DO. THEY JUST BLAST EVERYTHING LIKE A BOMB ESSENTIALLY. SO WE ARE WORKING ON ANTIBIOTICS THAT ONLY TARGET PATHOGENS AND NOT GOOD BACTERIA. YEAH, THAT'S SOMETHING THAT -- IT'S PART OF OUR EFFORTS IN TRYING TO ACHIEVE WHAT WE CALL TARGETTABILITY, WHICH IS ONLY WHAT YOU WANT AND NOTHING ELSE. THAT CAN MEAN YOU WANT TO TARGET TWO BACTERIA AT A TIME BUT NOT THESE OTHER THREE OR YOU WANT TO TARGET FIVE AND NOT THIS OTHER ONE. SO IT'S A VERY COMPLICATED PROBLEM BECAUSE OF COURSE YOU KNOW, WE DON'T KNOW ENOUGH ABOUT EACH BACTERIAL STRAIN IN TERMS OF PHYSIOLOGY, MEMBRANE COMPOSITION AND SO ON, SO IT'S A VERY HARD PROBLEM, BUT WE'RE TRYING OUR BEST. >>QUESTION ABOUT WHETHER OR NOT YOUR APPROACH IS AMABLE TO OTHEO OTHER DISEASES BESIDES ANTIBIOTIC RESISTANT INFECTIONS. >>THAT'S AN INTERESTING NOTION. I THINK OBVIOUSLY OUR FOCUS IS ON ANTIBIOTICS, BUT WE NOW HAVE A COLLABORATION TO DEVELOP ANTICANCER PEPTIDES, SO WE HAVEN'T VALIDATED THEM EXPERIMENTALLY YET SO I CAN'T REALLY SAY, BUT I DO THINK AT LEAST CONCEPTUALLY, SOME OF THE PIPELINES AND SOME OF THE APPROACHES WE'RE COMING UP WITH, I THINK THEY COULD POTENTIALLY BE EXTRAPOLATED TO OTHER INDICATIONS, NOT ONLY BACTERIAL INFECTIONS BUT ALSO MAYBE VIRAL INFECTIONS, MAYBE CANCER. WE'VE HAD PROJECTS IN THE PAST IN MALARIA AS WELL, SO I DO THINK SOME OF THESE THINGS MIGHT BE USEFUL IN OTHER AREAS. >>INTERESTING. ANOTHER CAREER QUESTION. SO THEY'RE A POSTDOC WITH A BACKGROUND IN COMPUTER SCIENCE AND MACHINE LEARNING, WANT TO KNOW WHAT YOU THINK THE NEXT STEPS WOULD BE TO PURSUE THE SORT OF RESEARCH YOU'RE DOING WITH SOMEONE WHO HAS A BACKGROUND IN THOSE AREAS. >>SO IF THIS IS TO THE POSTDOC, DEFINITELY REACH OUT TO LABS, TALK TO PEOPLE, AND YOU WANT TO MAKE SURE YOU ALIGN WITH THE LABS THEY ARE GOING TO. IF IT'S TO GO FOR THE NEXT STEP, FOR A FACULTY JOB OR A CAREER IN INDUSTRY, TRY TO FIND A GOOD PLACE. WHERE THE MISSION ALIGNS VERY MUCH WITH YOUR INTERESTS. IF YOU WANT TO DO COMPUTATIONAL WORK IN THE CONTEXT OF DRUG DISCOVERY, TRY TO GO TO A PLACE WHERE YOU'RE GOING TO BE HAPPY, WHERE YOU'RE GOING TO BE ALIGNING WELL AND DOING YOUR WORK EVERY DAY OR MOST DAYS. >>GREAT. A LOT OF THE PEPTIDES THAT YOU FOUND THE QUESTION UNDERSTOOD CORRECTLY SEEMED TO BE THE ONES IN THE GENOME ALREADY, SEEMED TO BE IN THE CONTEXT OFTEN OF A LARGER PROTEIN. SO THE QUESTION IS, DO YOU THINK THOSE PEPTIDES HAVE THE ANTIBIOTIC OR ANTIBACTERIAL FUNCTION IN THE CONTEXT OF THE LARGER PROTEIN OR IS IT JUST A COINCIDENCE THAT THEY HAPPEN TO HAVE THAT FUNCTION? >>YEAH, THIS IS A VERY GOOD QUESTION. SO ONE OF THE THINGS THAT I DON'T THINK I GOT INTO IT BUT SO A LOT OF THE ENCRYPTED PEPTIDES ARE ACTUALLY PREDICTED TO BE CLEAVED UP BY PROTEASES. >>INTERESTING. >>SO YOU HAVE YOUR ENTIRE PROTEIN, THEN A PROTEASE COMES ALONG PRESENT IN A PARTICULAR ENVIRONMENT IN THE BODY. IT CLEAVES OFF THE FRAGMENT AND THAT FRAGMENT IS WHAT HAS THE ANTIMICROBIAL ACTIVITY. SO THIS OBVIOUSLY SPARKS A LOT OF POTENTIAL THOUGHTS. ONE OF THEM IS THE NOTION OF GENOMIC -- PRESERVING ENERGY AT THE GENOMIC LEVEL WHERE ONE GENE ENCODES FOR ONE PROTEIN BUT THAT ONE PROTEIN HAS MULTIPLE FUNCTIONS. THAT CAN BE ACTIVATED BY AREAS CLEAVED BY PROTEASES, ESSENTIALLY -- PARTICULAR FRAGMENTS FROM A PROTEIN. SO PRESERVING ENERGY AT THE GENOMIC LEVEL. THINKING ABOUT PROTEINS NOT ONLY HAVING ONE FUNCTION BUT HAVING MULTIPLE FUNCTIONS THAT CAN BE ACTIVATED. >>WOULD THAT BE USEFUL FOR YOUR IDEA OF TARGETING THEM TO SPECIFIC AREAS SO IF THE MAIN PROTEIN CARRIED THEM TO THE RIGHT AREA AND THEN IT GOT CLEAVED OFF, IS THAT A STRATEGY? >>THAT MIGHT BE ONE OF THE THINGS THAT WE MIGHT EXPLORE. YOU CAN ALSO -- THE WAY I THINK ABOUT THIS, YOU HAVE A PROTEIN, DOING DAILY ROUTINE, RIGHT? A PROTEIN, FOR EXAMPLE, INVOLVED IN THE NERVOUS SYSTEM. UNDER PARTICULAR CONDITIONS, MAYBE AN INFECTION, I DON'T KNOW, THIS IS JUST SPECULATION, BUT MAYBE THE PERSON GETS INFECTED, A MICROENVIRONMENT WHERE PROTEASES GET RELEASED, CLEAVING OFF -- MAYBE DOZENS OF THEM IN A PARTICULAR AREA, AND THEN YOU HAVE ALL THESE COCKTAILS OF ENCRYPTED PEPTIDE ANTIBIOTICS THAT ARE WORKING -- THAT'S ME DREAMING A LITTLE BIT, IMAGINING THE SITUATION, BUT THAT COULD BE, YOU KNOW, POTENTIALLY WHAT MIGHT BE HAPPENING. >>VERY COOL. RELATED TO THE TARGETING IDEA, AND I HOPE YOU KNOW WHAT THIS QUESTION MEANS, WOULD FERRAL FLUIDS BE AN OPTION TO DIRECT ANTIBIOTICS TO A TARGET SITE? >>I'M NOT SURE ACTUALLY. I HAVEN'T REALLY EXPLORED THAT. THAT CONCEPT. FLEURN. .WOULD LOVE TO LEARN MORE. >>ON THAT NOTE, LEILA WOULD LIKE TO KNOW HOW TO GET IN TOUCH WITH YOU, IF SHE'S INTERESTED IN WORKING IN YOUR LAB IN THE FUTURE. SO I ASSUME VIA EMAIL? IS THAT THE BEST WAY? >>YEAH, I WOULD SAY EMAIL OR TWITTER OR -- EMAIL OR TWITTER. SO LET ME -- I CAN PUT IN THE CHAT. >>PUT IT IN THE CHAT, THAT WOULD BE GREAT. >>THAT'S MY EMAIL. THEN TWITTER HANDLE IS ?SH SO ANY OF THOSE TWO WORK VERY WELL. >>ALL RIGHT. WELL, I THINK THOSE ARE A LOT OF QUESTIONS, FANTASTIC QUESTIONS, BUT I THINK THAT'S WHAT WE HAD. SO CESAR, THANK YOU SO MUCH, THIS WAS A TERRIFIC LECTURE, TERRIFIC WORK, AND REALLY PLEASED TO SEE IT PROGRESSING SO WELL. >>JOHN, THANK YOU SO MUCH, AGAIN, FOR THE OPPORTUNITY, AND THIS HAS BEEN REALLY SUPER FUN AND A PRIVILEGE. >>THANK YOU SO MUCH. THANKS, EVERYONE, FOR JOINING. >>THANKS, EVERYBODY. TAKE CARE. >>BYE-BYE.