>> OKAY.
WHAT A NICE AUDIENCE.
I WOULD LIKE TO WELCOME YOU TO
TODAY'S NIH DIRECTOR LEKSURE
SERIES.
OUR SPEAKER IS DR. TERRY
PRZYTYCKA.
TERRY HAS A VERY INTERESTING
BACKGROUND.
SHE WAS A MATHEMATICIAN AS
UNDERGRAD AT WARSAU UNIVERSITY,
THEN GOT A Ph.D. IN COMPUTER
SCIENCE IN CANADA, UNIVERSITY OF
BRITISH COLUMBIA.
SUBSEQUENTLY WAS ON THE FACULTY,
AS ASSISTANT PROFESSOR AT
UNIVERSITY OF CALIFORNIA RIVER
SIDE, THEN MOVED TO THE
UNIVERSITY OF SOUTHERN DENMARK.
THEN CAME BACK TO THE UNITED
STATES WHERE SHE WAS IN THE
BIOPHYSICS DEPARTMENT AT
HOPKINS.
BEFORE WE WERE VERY FORTUNATE TO
RECRUIT HER TO COME TO THE NIH.
THE NLM NCBI COMPUTATIONAL
BIOLOGY BRANCH WHERE SHE'S
RECENTLY TENURED AS SENIOR
INVESTIGATOR.
SHE USES MATHEMATICAL MODELS TO
UNDERSTAND BIOLOGICAL PROCESSES.
THINKS FAIRLY BROADLY ABOUT
THESE THINGS, INCLUDING USE OF
GRAPH THEORY AND ALGORITHMIC
KINDS OF APPROACHES BUT NOT BY
ANY MEANS LIMITED TO THOSE.
JUDGING BY THE AUDIENCE, THERE'S
LOTS OF PEOPLE AT NIH INTERESTED
IN WHAT SHE DOES.
THE TALK TODAY IS CALLED TODAY'S
BRIDGING -- TOWARDS BRIDGING THE
GENOTYPE PHENOTYPE GAP, SYSTEMS
LEVEL DISSECTION OF IMPACT OF
VARIATION IN DNA SEQUENCE AND
STRUCTURE ON GENE EXPRESSION.
TERRY.
>> THANK YOU FOR THE
INTRODUCTION.
IT'S A PLEASURE TO BE SPEAKING
HERE TODAY TO SHARE WITH YOU THE
WORK THAT WE DO IN MY GROUP.
I WILL MENTION MY GROUP IS
PURELY COMPUTATIONAL SO WE APPLY
VARIOUS COMPUTATIONAL TECHNIQUES
THAT TO SOLVE QUESTIONS THAT
ARRIVE IN MICROBIOLOGY.
SO TODAY I HAVE A PROBLEM THAT
GENERALLY EVOLVES AROUND THE
QUESTION OF UNDERSTANDING THE
RELATIONSHIP WITH GENOTYPE AN
PHENOTYPE.
SO (INDISCERNIBLE) GENOTYPE
PHENOTYPE TYPICALLY STARTS WITH
ASSOCIATION STUDIES.
IN WHEN WE TRY TO IDENTIFY
GENETIC LOCATION IN THE GENOMES
VARIABILITY IN THIS PARTICULAR
LOCATION CORRELATES WITH WITH
THE PHENOTYPIC VARIABILITY.
WE CAN IDENTIFY A PLACE LIKE
THAT, IT -- SOME SUGGESTED THAT
IT IS IN THE GENERE GENERAL THIS
MIGHT BE A GENE THAT DIRECTLY OR
INDIRECTLY AFFECTS THE
PHENOTYPE.
SO THIS IS JUST A FIRST STEP, IN
ORDER TO UNDERSTAND THE
RELATIONSHIP BETTER WE NEED TO
GO DEEPER AND WE NEED TO
UNDERSTAND NOT ONLY TO IDENTIFY
THE ASSOCIATION BUT TO BE ABLE
TO POINT -- PINPOINT SPECIFIC
GENES OR REGULATORY ELEMENTS OR
SPECIFIC MUTATIONS THAT ARE
RESPONSIBLE FOR VARIATION.
WE NEED TO UNDERSTAND THE
MECHANISM OF ACTION, WHAT
EXACTLY IT DOES TO THE MUSCLE
IT'S EMBEDDED IN.
WE TRY ALSO TO UNDERSTAND HOW
THE AFFECT OF THIS ACTUAL
MUTATION OF THE PERTURBATION
AFFECT OTHER MOLECULES AND OTHER
PATHWAYS.
LAST BUT NOT LEAST WE HAVE TO
KEEP IN MIND THAT MOST OF THE
PHENOTYPES OF INFECTS, MOST OF
THE DISEASES, MOST ARE COMPLEX
PHENOTYPE AND WE SHOULDN'T
EXPECT THERE TO BE ONE TO ONE
RELATIONSHIP, JUST ONE GENETIC
PHENOTYPE.
INSTEAD COMPLEX DISEASES ARE
CHARACTERIZED AND COMPLEX
PHENOTYPE BY THE FACT THEY'RE
AFFECTED BY A NUMBER OF GENETIC
LOCATION, WHICH JOINTLY AFFECT
THE PHENOTYPE AND THOSE AFFECTS
DO INTERACT WITH OTHER EPISODIC
WAYS WHICH MAKES IT DIFFICULT TO
UNCOVER THOSE RELATIONSHIPS.
IN THE LAST FEW YEARS MY GROUP
HAS BEEN WORKING ON ALL THOSE
PROBLEMS AND ONE THING THAT'S
IMPORTANT TO KEEP IN MIND TO ASK
A DIVERSE QUESTION, WE NEED TO
LOOK AT THE MOLECULAR SYSTEM
WITH VARIOUS LEVEL OBSTRUCTION
AND VARIOUS LEVEL OF DETAIL.
SO TODAY I ORGANIZE MY
PRESENTATION WHILE LOOKING AT
THE QUESTION OF OF CLOSING
ENGINE TYPE PHENOTYPE PATHWAY.
ON A DIFFERENT LEVEL WE LOOK AT
THE STRUCTURE AND SEQUENCE
LEVEL, WE TRY TO UNDERSTAND HOW
THEY THE DYNAMICS ARE MOST
DIRECT CHANGES AND I'LL SAY A
LITTLE BIT ABOUT OUR WORK THAT
RELATES TO CONFIRMATIONAL
CHANGENING DNA STRUCTURE AN RNA
STRUCTURE.
THEN WE WILL NEED TO UNDERSTAND
HOW THE PERTURBATIONS PROPAGATE
THROUGH THE SYSTEM, WE TAKE A
NETWORK LEVEL APPROACH AND HERE
I'M GOING TO DISCUSS OUR
(INDISCERNIBLE) TO START
PROPAGATION OF THE VARIATION,
GENETIC VARIATION THROUGH THE
SYSTEM IN THE CONTEXT OF CANCER
HOW DOES VARIATION DISREGULATE
SPECIFIC MOLECULAR PATHWAYS AND
HOW IT CAN USE COMPUTATIONAL
APPROACHES TO SUGGEST OR PREDICT
THE GENES MUTATIONS THAT ARE
CAUSAL FOR THOSE DISREGULATIONS.
FINALLY I WILL RESUME FARTHER
TAKE ASSOCIATION TYPE APPROACH,
SHOW HOW WE CAN USING
ASSOCIATION UNCOVER EPISODIC
INTERACTION WHICH CAN GIVE US
QUITE INTERESTING INSIGHT INTO
BIOLOGY.
I WILL DISCUSS THIS IN THE
CONTEXT OF EPISTATIC INTERACTION
RELATED TO DRUG RESISTANCE.
I WOULD LIKE TO START WITH
CONFIRMATIONAL CHANGES IN DNA
STRUCTURE.
IN COLLABORATION WITH DAVID
LEVIN'S GROUP IN NCI AND
(INDISCERNIBLE).
I WOULD LIKE TO START BY
ACKNOWLEDGING THE COMPUTATIONAL
PART OF THIS WORK AND ALSO FOR
THE RESPECTIVELY WHO DID THE
EXPERIMENTS.
THE CONONCAL STRUCTURE OF DNA IS
THE DOUBLE HELIX.
HOWEVER, THERE'S AMPLE EVIDENCE
IN VITRO THAT DNA MAY ALSO
ASSUME AUTOMATIC STRUCTURES AS
THE MOST PROMINENT ARE QUAW QUAW
SINGLE STRANDED DNA.
BUT ALSO A NUMBER OF OTHER
STRUCTURES THAT CAN BE ASSUMED
BY DNA.
NOW, SO THOSE THE STRENGTH OF
EVIDENCE IN IN VITRO BUT ALSO IN
VIVO THERE ARE SOME INDICATIONS
AND -- THAT THOSE CONFIRMATIONAL
CHANGES MAY HAVE REGULATORY
RESULTS.
THEY MAY IMPACT THE GENES THAT
ARE ALMOST IN THE NEIGHBORHOOD
OF THE STRUCTURE AS PROPOSED
BYUTHEM 12 OR AS DOCUMENTED BY
THE PAPER DR. DAVID'S GROUP THEY
CAN ACTUALLY AFFECT GENES THAT
ARE AS FAR AS 1 KB FROM THE
PERTURBATION.
SO OF THOSE ISOLATED EXAMPLES
THIS IS SOME TYPE OF REGULATION
MORE BROADLY TO -- SO IN ORDER
TO UNDERSTAND WHETHER IT WAS
THIS TYPE OF ONE HAS TO SOMEHOW
BE ABLE TO IDENTIFY THOSE
STRUCTURES IN VIVO.
SO IN ORDER TO DO THIS, WE USE
THE EXPERIMENTAL GROUP IN DAVID
LEVIN'S LAB WHICH CAPPIZE ON
OBSERVATION OF ALL OF THOSE
STRUCTURES ARE ACCOMPANIED WITH
SOME BASIS.
THE PROCEDURE IS TO UNCOVER
THOSE BASES, WE REFER TO AS
SINGLE STANDARD -- SINGLE
STRANDED DNA 6 TO IMPORTANT PART
FOR US IS TO UNDERSTAND THAT AS
THE (INAUDIBLE) RECOGNIZING
UNPAIRED BASIS IN VIVO IS THE
FIRST STEP.
LAST STEP IS TO SEQUENCE
FRAGMENTS ADJACENT TO THE UNPAIR
BASIS AND MARK THEM TO REFERENCE
GENES.
NOW, THIS TELLS -- SOMETHING
THAT TELLS US WHAT IRPAIR BASIS
WERE BUT IT DOESN'T TELL US
WHETHER THOSE HAVE ANYTHING TO
DO WITH THE AUTOMATIC STRUCTURE.
HERE IS WHERE COMPUTATIONAL
COMPONENTS COME IN AN THOSE
STRUCTURES CANNOT BE FORM IN IN
VITRO IN DNA BUT THE STEP
SEQUENCE MOTIF OR CERTAIN
PROPERTIES OF DNA THAT ALLOW
FORMATION OF THE STRUCTURES.
THOSE SEQUENCE DEPENDENT
PROPERTIES CAN BE IDENTIFIED
COMPUTATIONALLY.
SO WITH THIS WE CAN PREDICT
PREDICTABLY WHERE DNA SEQUENCE
MAY FORM A STRUCTURE WITH
EXPERIMENTAL DATA.
>> YOU CAN LOOKING AT OVERLAP
BETWEEN EXPERIMENTAL DATA AND
THE PREDICTION, YOU CAN SEE IN
THE ISLAND ARE IN REACH IN
RESTING STATE.
AS HE EXPECTED IT BECOMES
PRELIMINARY ACTIVITY LEAVE SOME
UNPAIRED BASIS, LESS OBVIOUSLY
BUT IMPORTANTLY ALSO SO THAT
GIVES US SOME COMPLIMENT ON
UNPAIRED BASIS RELATED TO THIS
AUTOMATIC STRUCTURE.
HOWEVER WE WANTED TO HAVE A
LITTLE BIT STRONGER EVIDENCE,
THOUGH IT'S TRUE ALL OF THOSE
STRUCTURES DO -- ARE CONFIDENT
WITH THOSE UNPAIRED BASIS BUT
THE CONTEXT UNPAIRED BASIS OCCUR
DIFFERENT FOR EACH STRUCTURES.
HERE WE HAVE TWO SMALL OPENINGS
SEPARATED BY THE RIGHT HANDED
HELIX AND HERE WE HAVE MORE
COMPLICATED STRUCTURES.
SINCE THOSE OPENINGS ARE
DIFFERENT CONTEXT YOU SHOULD SEE
DIFFERENT SEQUENCE PATTERN WHEN
WE MAP ADJACENT FRAGMENTS.
HERE IS THE THERAPEUTIC
CONSIDERATION FOR THIS, WHAT WE
SHOULD EXPECT IS REMEMBER WE
SEQUENCED ADJACENT FRAGMENTS SO
WE SHOULD SEE ON BOTH ENDS, WE
SHALL SEE A PEAK OF FIVE HERE
FOLLOWED BY A PEAK OF THREE
PRIME STACK.
THEN IT SHOULD BE DEPLETION OF
TAK BECAUSE WE DONE HAVE FROM
THIS REGION AN PATTERN SO THIS
IS THE PICTURE WE WE SHOULD SEE.
OR IN PRACTICE THIS IS WHAT WE
SEE IN PRACTICE.
HOW ABOUT OTHER STRUCTURES?
THE B DNA, THIS IS RIGHT --
(INDISCERNIBLE) WE HAVE UNPAIRED
BASES BETWEEN IN THE TRANSITION
BETWEEN B TO C JUNCTION.
AGAIN, I'M NOT GOING TO REPEAT
THE EXERCISE WE DID BUT YOU CAN
AGAIN PREDICT WHAT WAS THE RISK
PATTERN AND COMPOUND WIT THE
EXPERIMENTAL DATA AND SEE THAT
IT REALLY AGREES.
SO WHAT I SHOULD MENTION, WHEN
WE LOOK AT -- AT PARTICULAR
SEQUENCE IN DNA WHICH IS --
WHICH HAS PREDICTED AUTOMATIC
STRUCTURE, WE DO NOT SEE TOO
MUCH -- THIS IS CONSISTENT WITH
THE UNDERSTANDING THOSE
STRUCTURES IF THEY EXIST ARE
REALLY -- TO CLARIFY THIS IS NOT
AROUND ONE SINGLE PREDICTED SITE
BUT INSTEAD PULL THE SITES
TOGETHER, ALIGN THEM IN THE
MIDDLE AND THIS IS A PROFILE OF
ALL THE PREDICTED FILES.
OKAY.
SO THIS KIND OF SUGGESTS THAT
THOSE ARE ENRICHMENT OF SINGLE
CELL DNA IN PREDICTED AND
NON-DNA STRUCTURE THAT
UNDERSTANDS THE DISTRIBUTION
WITH THE THERAPEUTIC PREDICTION,
THOSE STRONGLY SUGGEST ABUNDANT
OCCURRENCES OF NON-DNA
STRUCTURES IN VIVO.
SO THIS IS A WORK IN PROGRESS,
WE HAVE LOTS OF OTHER QUESTIONS
TO ANSWER, WHAT EXACTLY DOES IT
PLAY ROLE, HOW THEY INTERACT
WITH OTHER ELEMENTS, HOW DO THEY
INTERACT WITH -- DEPENDS ON CELL
TYPE AN CELL STATE, THOSE ARE
ALL QUESTIONS THAT WE NEED TO
ADDRESS.
NOW, LET ME SWITCH FROM -- GEARS
AND MOVE FROM CONFIRMATIONAL
CHANGES IN DNA STRUCTURE TO
CONFIRMATIONAL CHANGES IN RNA
STRUCTURES.
(INDISCERNIBLE) WHO HAS BEEN
LEADING THIS PART OF OUR WORK.
SO IN THIS PARTICULAR CASE WE
WERE VERY INTERESTED IN
UNDERSTANDING HOW SINGLE
NUCLEOTIDE POLYMORPHISM MIGHT
AFFECT MESSENGER RNA STRUCTURE.
WHY IS THIS INTERESTING?
IT'S QUITE WELL DOCUMENTED THAT
MESSENGER RNA STRUCTURE AND FIVE
PRIME UTR MIGHT HAVE A
REGULATORY ROLE.
IT MAY AFFECT DEPRIVATION RATE,
IT MAY AFFECT THE TRANSLATION
INITIATION, BUT ALSO STRUCTURAL
CHANGES MAY ALSO BE IMPORTANT AS
WE PROPOSE THAT THEY MAY HAVE
IMPACT ON SPLICING AND CAN
CHANGE THE KINETICS OF
TRANSLATION WHICH IN THEORY
COULD LEAD TO CHANGES IN
KINETICS OF FOLDING OF THE
PROTEIN CHAIN AND PERHAPS
POTENTIAL PROTEIN FOLDING.
SO WE WANTED TO DEVELOP A
(INAUDIBLE) THAT ALLOW US TO
ASSESS STRUCTURAL DIFFERENCES
THAT ARE BY SINGLE NUCLEOTIDE
POLYMORPHISM.
PERHAPS YOU'RE AWARE THAT THERE
ARE WELL ESTABLISHED
COMPUTATIONAL PROGRAMS THAT
PREDICT A MINIMUM OF MESSENGER
RNA.
SO WHY DON'T WE TAKE THOSE
PROGRAMS, NATIVE SEQUENCE
MINIMAL FOR THE -- AND SEE WHAT
ARE THE DIFFERENCE.
WE DECIDED THIS IS NOT A CORRECT
WAY TO APPROACH THIS PROBLEM.
AND DEVELOPMENT IS NOT CORRECT
BECAUSE THOSE THINGS ARE SO
SIMILAR TO EACH OTHER.
SO FROM THE PROSPECT, I WILL TRY
TO CONVINCE YOU FROM A PRACTICAL
PERSPECTIVE YOU HAVE TO TAKE A
VIEW OF THE ASSEMBLY OF
FACTORSCH YOU LOOK AT THIS
PROTEIN -- AS THE DNA SEQUENCE
-- MESSENGER RNA SEQUENCE AND
THIS SEQUENCE MAY ACTUALLY
ASSUME A NUMBER OF FACTORS AND
HERE IS KIND OF THIS STRUCTURE
VISUALIZED BY THIS CAN
(INDISCERNIBLE) THE PROBABILITY
OF OF ASSUMING STRUCTURES
DIRECTLY DEFINED BY THE THE FREE
ENERGY OF THE STRUCTURE, AND AT
A MINIMUM WILL BE ASSUMED WITH
THE HIGHEST PROBABILITY.
BUT OTHER FACTORS MIGHT BE
ASSUMED THAT'S CORRELATED TO THE
MINIMUM.
COMPUTATIONAL EXPERIMENTS SHOW
WHEN YOU LOOK AT THE STRUCTURE
AND THE NEXT BEST STRUCTURE, THE
ENERGY DIFFERENCE IS REALLY VERY
SMALL.
SUGGESTING NOT ONLY MINIMUM
STRUCTURES LIKELY BUT ALSO
STRUCTURES SHOULD BE CONSIDERED.
SO NOW WE HAVE TO NOT JUST
COMPARE TO A MINIMAL STRUCTURE
BUT ASSEMBLY OF STRUCTURESCH
THAT'S NEW SO WE NEED TO COME UP
WITH A WAY OF MEASURING THE
DISCOUNT BETWEEN TWO ASSEMBLY
STRUCTURES AND ALSO COCOME UP
WITH THE COMPUTATIONAL AND
COMPUTE THE DISTANCE.
NOW IT IS NOT HARD TO COME UP
WITH A MEASURE, I WON'T GO INTO
THE -- WE CAN BORROW
INFLAMMATION THEORY WHERE
(INDISCERNIBLE) IS USED TO
MEASURE THE DIFFERENT
DISTRIBUTION.
WE CAN LOOK ON BIOPHYSICS AND
USE DEVELOPMENT -- AND ZOOM IN
ON THE STRUCTURE AND LOOK AT
EVERY BASE PAIR AND HOW THE
PROBABILITY OF FORMING THIS IS
DIFFERENT IN ONE ASSEMBLY VERSUS
THE OTHER ASSEMBLY.
SO UNFORTUNATELY IT'S EASIER TO
COME WITH THOSE MATHEMATICAL
EQUATIONS AND TO COMPUTE THOSE
EFFICIENTLY.
SO THE IMPORTANT CONTRIBUTION
HERE THAT (INAUDIBLE) IS TO COME
UP WITH WHAT IS KNOWN IN THE
SCIENCE AND PROGRAMMATIC
APPROACH.
WHENEVER YOU CAN DESCRIBE A
PROBLEM USING A DYNAMIC
PROGRAMMATIC APPROACH THAT
USUALLY IMPLIES A SUFFICIENT
ALGORITHM TO ANSWER THE
QUESTION.
SO NOW WE HAVE DESIGNING PROGRAM
ALGORITHM TO MEASURE THOSE
DIFFERENCES AND IMPORTANTLY,
THOSE ARE NOT REDUNDANT
MEASURES.
THEY CAPTURE THE DIFFERENT
ASPECTS OF THESE DIFFERENCES SO
E WANTED TO SEE -- TO LOOK AT
THE DISEASE ASSOCIATED SNPs
AND SEE FOR DISEASE ASSOCIATED
SNPS WHETHER WE SEE SIGNIFICANT
CHANGES IN MESSENGER RNA
STRUCTURES.
WE USELYZE THE DATABASE --
UTILIZE THE DATABASE OF THE SNP
IN FIVE PRIME UTR REGION TO SEE
HOW MANY ARE ACCOMPANIED WITH
SIGNIFICANT CHANGES IN MESSENGER
RNA STRUCTURE IN THAT REGION.
THERE'S ACTUALLY LOT UP HERE, I
SAW THE BIGGEST OFFENDER SORTED
BY RELATIVE ENTROPY, THIS IS A
LIST OF DISEASES AN MUTATIONS, I
WANTED TO POINT THE
(INDISCERNIBLE) SYNDROME IS
BECAUSE A THRESHOLD IS ALSO
ASSOCIATED WITH OCCURRENCES OF
THE NON-DNA STRUCTURE
SPECIFICALLY OF THE FORM.
THE SECOND THING WE WOULD LIKE
TO SEE CAN -- CHANGES IN
MESSENGER RNA EXPLAINED DISEASE
ASSOCIATED SNPS.
HERE WE'RE IN COLLABORATION WITH
TWO GROUPS LOOKING AT GENES
EXPERIMENTALLY CONSUMED TO HAVE
THOSE SILENT MUTATIONS THAT
AFFECT GENE FUNCTION SO THE
MUTATIONS IN THE REGION THAT
CHANGE IT IS PROBLEM BUT IT
DOESN'T CHANGE THE AKNEE KNOW
ACID SEQUENCE.
SO THE PROBLEM IS NOT -- AMINO
ACID SEQUENCE.
THE PROBLEM IS IN SOMETHING THAT
HAPPENS BEFORE, SO WE AND OUR
COLLABORATORS WOULD LIKE TO KNOW
THE MESSENGER RNA STRUCTURE
AROUND THE SNPs AND THE SHORT
ANSWER IS YES, THOSE ARE
SIGNIFICANT CHANGES IN MESSENGER
RNA STRUCTURE AT LEAST AROUND
SOME OF THOSE SNPS.
IN SUMMARY FOR THIS PART, WE HAD
EFFICIENT COMPUTATIONAL THAT
ALLOW US TO MEASURE THE NUMBER
OF SNP MESSENGER RNA STRUCTURE
THAT HELP US TO BREACH AN
IMPORTANT WAY THIS GAP BETWEEN
GENOTYPE AN PHENOTYPE.
IF WE HAVE A DISEASE CAUSING SNP
NOW WE CAN ASK THE NEXT QUESTION
WHETHER OR NOT IT'S POSSIBLE
THOSE ARE CHANGES IN MESSENGER
RNA THAT ARE CAUSATIVE FOR THE
UNDERLYING PERTURBATION.
NOW I WOULD LIKE TO ZOOM OUT AND
START TO TAKE THE NETWORK LEVEL
APPROACH AND ASK THE QUESTION
HOW DOES GENETIC PERTURBATIONS
LIKE THE ONE IN THE PREVIOUS
PART PROPAGATE THROUGH THE
SYSTEM?
ESPECIALLY IN THE CONTEXT OF
DISEASES HOW THIS REGULATES
PATHS.
THIS QUESTION WE LOOK AT
DIFFERENT TYPE OF GENETIC
VARIATION, MAINLY COPY NUMBER
VARIATION.
THIS IS NOT BECAUSE THE MATTER
IS NOT APPLIED TO OTHER GENETIC
VARIATION, BUT BECAUSE THIS IS
FOR THIS PARTICULAR VARIATION WE
HAVE DATA THAT IS BEST SUITED
AND YOU HAVE THE DATA AND YOU
HAVE A WONDERFUL COLLABORATION
TO LOOK AT THE COPY NUMBER
VARIATION TOGETHER.
SO COPY NUMBER VARIATION
ACTUALLY VERY IMPORTANT FOR
UNDERSTANDING DISEASES
INDEPENDENT OF MY PRACTICAL
POINTS THAT THIS IS WHERE WE
HAVE DATA.
NAMELY THAT A LOT OF UNDERLYING
DISEASES SUCH AS CROHN'S DISEASE
AND (INAUDIBLE), IT REVEALS MUCH
MORE STRUCTURAL VARIATION CALLED
FLEXOR VARIATION IN GENETICS.
SO QUITE A LOT OF THOSE
VARIATIONS IN THE POPULATION AND
PERHAPS SURPRISINGLY INCLUDING
COPY NUMBER VARIATIONS THAT
AFFECT WHOLE GENOME OR WHOLE
AXOMES.
SO THEY'RE QUITE FREQUENT AND
FREQUENT SOMATIC MUTATION IN
CANCER.
SO WE'LL START AGAIN WITH OUR
COLLABORATIVE WORK WITH BRAD
LEVER'S GROUP WHERE WE ARE
TRYING TO DISSECT THE IMPACT OF
GENE COPY NUMBER VARIATION ON
EXPRESSION OF GENES IN
(INDISCERNIBLE) FROM
(INDISCERNIBLE) GROUP WHO DID
ALL THE EXPERIMENTS AND A GREAT
DEAL OF ANALYSIS, THEY ARE DOING
ANALYSIS ESPECIALLY PART OF THAT
RELATES TO NETWORK.
WHAT YOU ARE DOING HERE, WE USE
POWER OF FLIGHT KINETICSCH THOSE
ARE LINES OF FLIGHT THAT THAT
ARE DEFICIENT IN CERTAIN DNA
REGIONS REMOVED.
AND WHENEVER YOU HAVE THOSE
REGIONS REMOVED, WHATEVER GENES
ARE SUPPOSED TO BE IN THE REGION
ARE NOW IN ONE COPY VERSUS IN
TWO COPIES.
SO WE HAVE THOSE FLYING COPY
NUMBER VARIATIONS.
SO IF THOSE HAVE FLIGHT IN
DIFFERENT REGIONS AND WE MOVE IN
OUR STUDY WE USE 21 OF THOSE
LINES, MUCH MORE IN LIBRARY.
SO WHAT YOU SEE@N$rQ)E ARE THIS
STRUCTURAL REPRESENTATION OF ALL
THE 21 LINES WITH ON CHROMOSOME
DELETION, ALL ON CHROMOSOME 2
AND YOU CAN SEE WHERE ON THE DOT
SHOW BLACK BARS TELLS YOU WHERE
THE DELETION IS.
SO THE FIRST QUESTION WE WANT TO
ASK IS WE REDUCE FOR THOSE GENES
WITH REDUCE COPY NUMBER FROM TWO
TO ONE, WHAT HAPPENS AS
EXPRESSION IN THOSE PARTICULAR
GENES.
HE WOULD EXPECT -- RETUESDAY THE
COPY NUMBERS BY HALF EXPRESSION
SHOULD DROP BY HALF.
THIS IS NOT PRECISELY WHAT'S
HAPPENING THOUGH FOR MOST GENES
EXPRESSION DROPS.
WHAT YOU SEE HERE IS A GRAPH, WE
PLOT THE CHANGE OF EXPRESSION A
LOG SCALE WITH RESPECT TO NORMAL
VERSUS THE WILD TYPE.
IF EXPRESSION DROP BY HALF YOU
SHOULD SEE ALL THOSE DROP AROUND
MINUS ONE.
WE SEE SOME BUT CERTAINLY NOT
ALL.
WE SEE ACTUAL SPECTRUM OF THEM
INCLUDING CORRECT NUMBER OF
GENES WIDOW NOT CHANGE OR ALTER
EXPRESSION AND MORE SURPRISINGLY
ARE GENES WHICH DESPITE THE FACT
THAT EXPRESSION IS REDUCED, COPY
NUMBER IS REDUCED BY HALF,
EXPRESSION IS LARGER AND QUITE A
GOOD BIT WHICH GOES BELOW,
EXPRESSION GOES LOWER THAN
WHAT'S JUSTIFIED THAN BY
REDUCING THE NUMBER OF GENES THE
GENES BY HALF.
NOW, THIS IS A REL TALLY YOU MAY
THINK THIS IS NOT A LARGE NUMBER
OF GENES THAT UP AND BEYOND OR
JUST GO OUT OF CONTROL BOUNDS,
BUT THIS IS THE SAME PICTURE WE
ALSO SEE IN CANCER.
WHEN YOU LOOK AT CORRELATION OF
COPY NUMBER OF GENE EXPRESSION
THE CORRELATION IS POSITIVE A
GREAT DEAL GENES RELATION IS
SURPRISINGLY -- IF THE COPY
NUMBER GOES UP THE EXPRESSION
GOES DOWN.
SO WE WOULD LIKE TO UNDERSTAND
THAT DEEPER SO TO DO THAT WE PUT
INTO CONTEXT OF NETWORK.
HERE WE HAVE A NETWORK THAT IS
CONSTRUCTED BY NOT SO LONG AGO
AND WHAT WE'RE TRYING TO SEE IN
A MOMENT IS A MOVIE WHICH WILL
BE IT RATED OVER ALL THIS 21
DELETIONS AND FOR EACH DELETION
YOU SEE GENES OUTSIDE WHICH GO
UP OR DOWN.
I PUT THOSE CYCLES HERE SO THAT
WHEN YOU HAVE -- I WANTED TO
MAKE A POINT THAT THIS CHANGES
NON-RANDOM.
MAWNLY YOU SEE THE GENES THAT
ARE IN THE SAME NEIGHBORHOOD
TEND THE CHANGE IN TANDEM.
THEY TEND TO BE UP OR DOWN OR
AUTOMATIC.
SO THAT SUGGESTS STRONGLY THE
CHANGES WE -- THE WAY THOSE COPY
NUMBER VARIATIONS AFFECT OTHER
GENES ARE DEPENDENT ON THE
NETWORK THAT WE HAVE IN HAND.
SO REALLY WOULD LIKE TO
UNDERSTAND THIS RELATIONSHIP
BETWEEN COPY NUMBER VARIATION
AND HOW THOSE COPY NUMBER
VARIATION PROPAGATE TO AFFECT
EXPRESSION OF OTHER GENES IN THE
NETWORK AND LOTS OF GENES
OUTSIDE WITH SOME PARTICULAR
GROUP AFFECT THOSE GENES WHICH
GO UP DESPITE THE FACT THAT WE
HAVE THE NUMBER OF GENE DOSAGE.
AS YOU REMEMBER 21 EMPERIMENTS
SO WE CAN'T GO LOOK AT WHOLE
NETWORK BEFORE WE DECIDED TO
LOOK JUST AS THE FIRST
NEIGHBORS.
SO WE DEFINE THIS AS A GENE INTO
THOSE THREE GROUPS AND LOOK AT
THE FIRST NEIGHBOR OF ALL OF
THOSE GENES IN THE NETWORK AND
SEE WHAT THE FIRST NEIGHBOR OF
ANY OF THOSE GROUPS CHANGE MORE
THAN OTHER GROUP.
WE FOUND FOR THE EXTREME GROUPS
IN MANY DIFFERENT WAYS, THE
NEIGHBOR OF THOSE GENES THAT ARE
GOING OUT AND BEYOND HAVE MORE
CHANGES THAN ALL THE OTHER
GENES.
SUGGESTING THERE IS A
PROPAGATION OF THE SYSTEM FROM
THOSE GENES TO THE NETWORK AND
FEEDBACK FROM THE NETWORK TO
THOSE GENES.
AND THIS IS BASICALLY WHAT I
SAID.
COPY NUMBER EXPRESSION INk>x A
NETWORK DEPENDENT WAY.
NETWORK GUIDE THOSE
PERTURBATIONS.
AND ALSO WE HAVE INDICATION OF
THAT THOSE GENES THAT ARE IN THE
NETWORK ACT ON THE DELETED GENES
TOO.
NOW I THINK THIS IS MOSTLY
MOTIVATION FOR ME TO GO TO THE
NEXT -- TO JUSTIFY NEEDS TO
DEVELOP AN APPROACH THAT IS
GOING TO MODEL HOW THE
PERTURBATION IN GENETIC MAY
PROPAGATE THROUGH THE SYSTEM TO
AFFECT OTHER GENES.
WE ARE AFFECTED THOUGH THE
APPROACH IS VERY GENERAL WE
DEVELOP IN THE CONTEXT OF CANCER
WHERE A NUMBER OF COPY NUMBER
VARIATION AND THE DATA IS
SUFFICIENT FOR US TO GO AND TEST
THE APPROACH.
AND I'LL STOP BY ACKNOWLEDGING
(INAUDIBLE) WHO HAS BEEN HELPING
THIS PART OF OUR WORK.
SO MOTIVATION OF COURSE WE WOULD
LIKE TO HAVE TO ANSWER THE
QUESTION I JUST POSTED WITH THE
BIGGEST PART BUT ALSO IN THE
CONTEXT OF DISEASE WOULD LIKE TO
IDENTIFY PATHWAYS OF
DISREGULATED IN DISEASES.
NOT WELL UNDERSTAND THAT COMPLEX
DISEASES SHOULD BE DISTORTED
FROM THE CONTEXT OF PATHWAYS,
THERE MAYBE VARIATIONS WHICH END
UP DISREGULATING THE PATHWAYS
LEADING TO THE FACT THAT THERE
HAVE BEEN A DIFFERENT CAUSAL
MUTATION THAT IN THE END WILL
GIVE THE SAME OR RELATED TYPE OF
TUMOR.
SO WOULD ALSO LIKE THE CAPTURE
THE MOST IMPORTANT PLAYERS IN
THIS NETWORK, PERHAPS NETWORKS
IN THE DISEASES THAT HELP TO
SUGGEST DEVELOPMENT ALSO WOULD
LIKE TO BE ABLE TO IDENTIFY MANY
MUTATIONS THAT WE HAVE IN SOME
CELLS WHICH MUTATIONS ARE MOST
PROMINENT, MOST IMPORTANT OR
WHICH HAVE THE DRIVING MUTATIONS
FOR THE CHANGES EXPRESSION
CHANGES THAT WE OBSERVE.
O HERE IS THE DATA WE HAVE A SET
OF CONSULTATION DATA FROM
COMPOSITION WHICH CONTAINS TWO
TYPE OF DATA.
EXPRESSION DATA AND THOSE
EXPRESSION ARE COMPARED TO
NON-CANCER SO WE KNOW WHICH
GENES ARE OVEROR UNDER
EXPRESSED, AT THE SAME TIME WE
HAVE COPY NUMBER VARIATION FROM
THE SAME SET OF PATIENTS.
THE QUESTION IS HOW THE
PERTURBATIONS THAT STARTED FROM
COPY NUMBER VARIATION PROPAGATE
THROUGH THE SYSTEM TO AFFECT THE
EXPRESSION OF THE GENES.
SO WE WOULD LIKE TO PUT IT IN
THE CONTEXT OF THE NETWORK AND
WE INTEGRATE PROTEIN PROTEIN,
PROTEIN DNA AN PHOSPHORYLATION
NETWORK THAT ARE OBTAINED FROM
HIGH THROUGH PUT EXPERIMENTS.
BEFORE I -- HOWEVER, LET ME TAKE
THE NETWORK AWAY AND LET ME
REMIND YOU ONE OF THE SLIDES
WHEN I SAID THAT OUR FIRST STEP
UNCOVERING THE RELATIONSHIP
BETWEEN GENOTYPE AND PHENOTYPE
ARE DISSOCIATION STUDIESCH GENE
EXPRESSIONS ARE CONSIDERED AS A
PHENOTYPE.
SO THE FIRST STEP WE WOULD LIKE
TO MAKE IS TO FOR EACH OF THOSE
GENES THAT ARE EXPRESSED
ABNORMALLY TO SEE WHETHER
EXPRESSION OF THESE GENES IS
CORRELATED WITH COPY NUMBER
VARIATION.
SO WE TAKE GENE EXPRESSION
PHENOTYPE AND THIS IS EXACTLY
WHAT IS KNOWN AS ANALYSIS,
EXPRESSION ANALYSIS AND ANAL
SITS IS KNOWN TO HAVE IMPORTANT
PROBLEMS.
AND THE IMPORTANT PROBLEM IS
THAT WE HAVE THOUSANDS OF GENES,
WE HAVE THOUSANDS OF VARIATIONS
THEN WE WILL TEST PAIRS AND WE
DON'T HAVE ALL THE SAMPLES TO
REALLY RECOVER FROM THE MASSIVE
MULTIPLE THE TESTING ISSUE WHEN
WE DO THAT.
WE CANNOT GET AND COMPARE SO WE
HAVE TO BE SMARTER ABOUT THAT.
SO WHAT WE DECIDED THE FIRST
STEP WE'RE GOING TO FIND TARGET
GENES.
GENES EXPRESSIONED AND CONTROL
WHICH ARE MARKERS OF THE DISEASE
OR WHICH ARE REPRESENTED --
REPETITIVE OF THESE DISEASES.
IN ORDER TO IDENTIFY THE SET OF
GENES WE FORMALIZE THE QUESTION
AS AN OPTIMIZATION QUESTION AND
WE REPRESENTED THE RELATIONSHIP
USING GRAPH THEORY, THIS IS
SOMETHING I WOULD LIKE TO SKIP
OVER IN THE INTEREST OF TIME AND
JUST SUMMARIZE WE USE OUR
TOOLBOX TO COME UP WITH A SET OF
GENES THAT REPRESENT THE
DISEASES, SO THAT WE ARE
CONFIDENT THAT IF WE EXPLAIN THE
CHANGES OF THOSE GENES, WE WILL
GAIN QUITE A LOT OF
UNDERSTANDING OF WHAT'S
CHANGING.
SO NOW WE HAVE A SMALLER SET,
MUCH SMALLER SET AND NOW WE CAN
AFFORD TO DO DISSOCIATION AND
SOME OF THOSE MIGHT TURN OUT TO
BE SIGNIFICANT.
SO UP HERE MARKED BY -- WHENEVER
YOU HAVE A SIGNIFICANT
ASSOCIATION BETWEEN COPY NUMBER
VARIATION AN EXPRESSION
VARIATION, WE KNOW THIS IS
SUGGESTED IN THIS REGION,
SOMETHING IN THIS REGION, ONE OF
THOSE GENES THERE MAYBE HUNDREDS
OF THEM IN THIS REGION THAT
AFFECT DIRECTLY OR INDIRECTLY
THIS TARGET GENE HERE.
SO NOW IT IS TIME FOR US TO PUT
OUR NETWORK I WOULD NOW LIKE TO
EXPLAIN HOW IT'S POSSIBLE THAT A
SIGNAL TRAVELS FROM ONE OF THOSE
GENES, WE SIMPLIFY THE QUESTION
AND ASSUME THAT THIS IS ONE OF
THOSE GENES.
SO HOW IT'S POSSIBLE FROM ONE OF
THE GENES THAT ARE IN THIS
REGION WITH COPY NUMBER
VARIATION TO THE TARGET GENE.
NOW, AS FAR AS AS THOSE HERE WE
-- THE NETWORK THAT WE HIGH
THROUGH PUT MATTER OBTAINED OUT
OF CONTEXT HAS BEEN OBTAINED IN
VITRO SO IT'S BUT IN VIVO IT'S
NOT IN THE (INDISCERNIBLE) WE
STUDIED.
SO IT IS SOMETHING BUT WE CANNOT
RELATE EXACTLY THOSE, WE NEED TO
SOMEHOW PUT IT IN THE CONTEXT OF
OUR PARTICULAR DISEASE, IN THE
CONTEXT OF CANCER.
FORTUNATELY WE HAVE THE
EXPRESSION DATA FOR ALL GENES
INCLUDING THE GENES IN THE
NETWORK.
SO WE CAN USE IT THEN WE REASON
MANY WAYS IN WHICH THOSE GENES
CAN CONNECT TO THESE GENES, THE
PATHWAYS THAT EXPRESSION OF
THOSE CORRELATE WITH EXPRESSION
OF TARGET GENES IS MORE LIKELY
TO (INAUDIBLE).
SO WE LIKE TO FIND A PATHWAY
THAT CONNECTS THIS REGION WITH
THIS REGION CHOOSING THAT ARE
CORRELATED WITH OUR TARGET
GENES.
TO BIAS OTHER TYPE OF PATHWAY WE
REPRESENT ALL THESE NETWORK
(INDISCERNIBLE).
SO THIS ALLOW US TO PUT OUR
RESISTANCE ON THIS AND WHENEVER
THE RESISTANCE ARE BEING DEFINED
BY HOW WELL THE TWO ADJACENT
NODES, EXPRESSION OF TWO
ADJACENT NODES CORRELATE WITH
EXPRESSION OF THE TARGET GENE.
SO IN THIS WAY WE CAN BIAS THE
CURRENT TO FLOW THROUGH THE
LOWEST RESISTANCE PATH, DOESN'T
HAVE TO BE JUST -- DOESN'T HAVE
TO BE A LINEAR PATH, MAYBE A SUB
NETWORK REALLY.
BUT NOT ONLY CAN WITH FIND THE
NODES THAT FACILITATE THE
CONNECTIONS BETWEEN THIS AND
THIS REGION, BUT ALSO BY THIS
APPROACH IDENTIFY ONE OF MANY
GENES THAT MAY HAPPEN AND
SOMETIMES OVER HUNDREDS OF THEM
THAT MAYBE IN THIS ENGINE LOCUST
WHICH IS MORE LIKELY CAUSAL FOR
THE VARIATION.
SO WITH THIS APPROACH SO FAR, WE
HAVE REALLY WE CAN PUT HANDS ON
THREE INTERESTING SET OF GENES.
THE FIRST SET IS THE SET OF OUR
MARKER GENE, THE GENES THAT ARE
WE SELECTED AS REPRESENTATIVE
GENES FOR DISEASES.
THE LIST OF THIS GENE IF YOU ARE
DOING CANCER RESEARCH AND
PROBABLY FIND YOUR GENES ON THIS
LIST.
JUST TO MENTION THE RAS ARE
GENES THAT ARE UP REGULATED AND
DOWN REGULATED.
THE SECOND SET IS THE CAUSAL
GENES.
THE GENES WITH COPY NUMBER
VARIATION ARE MORE LIKELY TO
CODE THE PERTURBATION OF THOSE
GENES MEANING THERE'S
SIGNIFICANT CURRENT FROM THOSE
GENES TO THOSE GENES.
AGAIN HERE THE GENES WHICH ARE
AMPLIFIED BY COPY NUMBER
VARIATION, THESE ARE THE GENES
THAT ARE DELETED.
IN BLUE ARE SOME KNOWN GENES
THAT ARE KNOWN TO BE CRUCIAL FOR
GLIOMA.
IN THE MOUSE WE CAN CAPTURE THE
NULL THAT BELONGS TO MANY
PATHWAYS, NETWORK AND AGAIN
AMPLIFIED IN THE -- OVER
EXPRESSED IN THE KAREN GREEN ARE
UNDEREXPRESSIONED IN THE CANCER
AND THAT HAPPENS TO BE IN OUR
NETWORK BUT IF IF YOU LOOK AT
THE EXPRESSION YOU WOULDN'T SEE
A CLEAR SIGNATURE THAT THEY ARE
ABNORMAL OTHERWISE.
AND AGAIN YOU RECOGNIZE KNOWN
CANCER GENES AND LEADING THE WAY
IS WITH THE LAST 100 PATHWAYS
HERE.
SO THESE ARE THE GENES IN THOSE
PATHWAYS.
WE'RE GOING TO IDENTIFY PATHWAYS
THAT ARE DISREGULATED BY THOSE
COPY NUMBER VARIATIONS.
TO DO SO, REMEMBER WE HAVE NOW A
LOT OF PAIRS THAT WE IDENTIFY
THAT COPY EXPRESSION OF THIS
PACKAGE GENE IS CORRELATED WITH
THE COPY NUMBER VARIATION OF
THIS REGION SO WE HAVE LOTS OF
THOSE PAIRS.
EAR ARE TWO OF.
FOR EACH OF THOSE PAIRS FOR SOME
OF THEM WE COULDN'T FIND
STATISTICALLY SIGNIFICANT
CURRENT GOING HERE AND THERE AND
THEN WE DROP THEM.
SO FOR MANY OF THESE PAIRS WE
COULD FIND STATISTICALLY
SIGNIFICANT CURRENT FLOW BETWEEN
THIS REGION AND THIS GENE AN THE
DOTS HERE REPRESENT THE GENES
THAT ARE SIGNIFICANTLY AFFECTED
BY THIS CURRENT.
I DROP THIS EDGE BECAUSE REALLY
WE WOULD LIKE TO THINK ABOUT
THOSE GENES.
WE HAVE TO KEEP IN MIND THAT OUR
NETWORK ARE VERY NOISY AND WHY
IN GENERAL WE HAVE A GOOD
INDICATION THAT THE GENES THAT
ARE CLOSE IN THE NETWORK ARE
RELATED, I WOULD BE NOT SO GREAT
TO BE SURE THAT THIS EXACTLY --
THE SIGNAL BIOLOGICAL SIGNAL
EXACTLY IS REPRESENTED BY THE
CURRENT.
SO I'M LOOKING AT THEM FROM THE
PERSPECTIVE OF THE GENES AND I
I'M ASKING THE QUESTION IF I
LOOK AT THOSE BACK UP GENES WHAT
IS THE INTERSECTION?
THE GENES IN THE INTERSECTION,
IF THEY BELONG TO SOME SPECIFIC
BROAD PATHWAY, WE WILL GET A
HOLD OF WHICH PATHWAYS ARE
REALLY OFTEN DISREGULATED BY
THOSE DIFFERENT MUTATIONS.
HERE ARE THE LIST OF THOSE
PATHWAYS THAT WE DISCOVERED,
PATHWAYS THAT ARE IN THE OVERLAP
AND TO DEFINE THE PATHWAY WE
USED GENE ONTOLOGY ANNOTATION
USING THE THE MOST SPECIFIC TERM
THAT WE CAN APPLY FOR A GIVEN
SET, YOU COULD SEE THAT CELL
CYCLE IN OVERLAP FOR EPIDEMIC
GROWTH FACTOR RECEPTOR WHICH WAS
NINE AND SO SO ON AS THERE WAS A
LIST HERE OF ALL THE PATHWAYS
THAT ARE PURE, MORE THAN ONE IN
MORE THAN ONE PATHWAY, MORE THAN
THE BACK UP GENES SO AGAIN A
REALLY -- AS YOU LOOK AT THE
PATHWAYS, NOT SURPRISED TO SEE
THEM BE INVOLVED IN CANCER, SOME
OF THEM ARE PERHAPS MORE
SURPRISING AND THOSE PROBABLY
ARE MORE INTERESTING.
THEY MAY ALSO BE POSITIVE TOO.
SO WE HAVE A NUMBER OF
ADVANTAGES TO THIS APPROACH.
FIRST OF ALL IMPORTANT THAT WE
COULD USE EXPRESSION ENTERACT ON
DATA TO PUT SOME CONTEXT, TO PUT
THE CONTEXT OF THE DISEASE WITH
WHICH WE ARE WORKING.
CURRENT FLOW IS -- THIS IS A
BALANCING OF THE EQUATION THAT
WE CAN SOLVE E FUSHTLY.
THOSE ARE HUGELY EQUATIONS,
NUMBER OF VARIABLES IS EQUAL TO
THE NUMBER OF GENES SO IT'S NOT
(INDISCERNIBLE) PROBLEM AND WE
ALSO NEED TO SAY WHETHER THE
CURRENT WAS STATISTICALLY
SIGNIFICANT OR NOT SO WE NEED TO
HAVE EXPERIMENTATION TESTS SO WE
SAW THIS SET OF (INDISCERNIBLE)
SO WE DEVELOP AUTOMATIC APPROACH
WHICH ALLOWS US TO REUSE PARTIAL
SOLUTION AS WE DO OUR
PERMUTATION TEST.
THIS IS NOW IN PRINT.
MANY OF THE ADDRESS THAT WE HAVE
IN THE NETWORK, THOSE ARE
PROTEIN DNA INTERACTIONS,
PHOSPHORYLATION, THEY SHOULD
REALLY BE DIRECTED.
AND WE EXTEND THE FIELD APPROACH
DIRECTED TWO WAYS OF DOING THAT,
YOU CAN PUT CONSTRAINT TO FORCE
IT INTO ONE DIRECTION OR HAVE A
HEURISTIC APPROACH.
SO I ACKNOWLEDGE THIS IS THE
FIRST APPROACH IN THIS PART OF
THE LITERATURE THAT IDENTIFY
PATHWAYS THAT ARE DISREGULATED
THAT CONNECTS GENOTYPE AND
PHENOTYPE AND THIS JUST CAME
COUPLE OF MONTHS AGO IN
COMPUTATIONAL BIOLOGY.
OKAY.
WE HAVE GOT -- MY LAST SEVEN
MINUTES I WILL USE TO GO EVEN
HIGHER AND OBSTRUCTION AND LOOK
AT THE INFORMATION THAT WE CAN
GAIN ABOUT THE MOLECULAR SYSTEM
ASSOCIATION INVOLVED.
HERE I'M GOING TO LOOK AT THE
COG MITIVE INTERACTION.
LET'S START BY DEFINING
EPISTATIC INTERACTION BECAUSE
THERE'S RAREIOUS WAYS OF
DEFINING THEM.
EPISTATIC INTERACTION WHAT I'M
GOING TO UNDERSTAND FOR THIS
PRESENTATION INTERACTION IS THE
SITUATION WHEN YOU HAVE TWO
GENES.
WHICH ARE ARE -- AFFECT THE SAME
PHENOTYPE BUT THAT AFFECT IS NOT
INDEPENDENT.
MEANING THAT WHATEVER AFFECT
THIS HAS ON THE PHENOTYPE --
GENOTYPE CAN BE MODIFIED BY THIS
PHENOTYPE.
THIS CAN BE EXPLAINED IN
SYMPATHETIC LOCALITY AND I
SWITCH TO THIS BECAUSE THE REST
OF MY TALK WILL REFER TO THIS
ORGANISM.
SO THE IDEA IS THAT WE DELETE A
GENE AND WE LOOK AT THE ORGANISM
CAN STILL GROW AND I ASSUME IT
CAN DO THAT.
IT STILL GROW.
WHEN YOU MOVE BOTH OF THEM, IT
DOESN'T.
SO THAT SUGGESTS THAT THERE IS
SOME HIDDEN RELATIONSHIP BETWEEN
THOSE TWO GENES BECAUSE IF THEY
WERE INDEPENDENT WE WOULD EXPECT
THAT IT SHOULD BE FINE IN BOTH
OF THEM.
SO THIS INTERACTION AS WE CALL
IT, EPISTATIC INTERACTION, AND
THAT PARTICULARLY IS IMPORTANT
IN THE STUDY OF DRUG RESISTANT
PHENOTYPE.
SO HERE IN THIS STUDY WHICH WAS
NOW ACKNOWLEDGE (INDISCERNIBLE)
IN MY GROUP.
WE USED GENETIC CROSSES TO
UNCOVER EPISTATIC INTERACTION IN
CONTEXT OF DRUG RESISTANCE.
SO THE IDEA IS WE HAVE TWO THAT
HAVE DIFFERENT GENETIC
BACKGROUNDS.
SO OUR APPROACH, WE ASSUME THAT
THEY HAVE THE SAME PHENOTYPE.
THEY WILL BE BOTH DRUG
RESISTANT.
WHEN WE CROSS THEM WE OBTAIN A
NUMBER OF PROGENIES, SOME MIGHT
BE DRUG RECESS STAN, SOME MIGHT
NOT.
SO WE HAVE THOSE TWO PARTS THAT
ARE RESISTANT.
THE GENOME FOR PART 1 AND THEN
FOR THE THE WE HAVE THE MOSAIC
OF THOSE PARENTAL STRAINS.
WE ALSO HAVE THE PROGENY, THE
PHENOTYPE WHICH IS MIX 1.
HOW THE EPISTATIC INTERACTION
MIGHT EXPLAIN THAT TYPE OF
SITUATION?
ONE EXPLANATION, THEY MAYBE MANY
EXPLANATIONS WUK ONLY FIND ONES
CONSISTENT WITH OUR MODEL.
O OUR MODELS FOLLOW, IT'S
POSSIBLE THAT WE HAVE TWO
GENOMIC LOCATIONS.
BUT THAT BY WII INTERACTS WITH
EACH OTHER WHICH THOSE
INTERACTIONS DEVELOPED
INDEPENDENTLY IN EACH OF THE
STRAINS.
WHENEVER WE HAVE BOTH OF THOSE
LOCATIONS INHERITED FROM THE
SAME PART, THEY DON'T HAVE THE
STANDARD DRUG.
ONE IS INHERITED FROM ONE
PARTNER, ANOTHER IS INHERITED ON
THE OTHER PART.
SO THEY MAY HAVE THE
COMPATIBILITY, THEY MAY SAY --
OBTAIN INDEPENDENTLY IN EACH
STRAIN AND INTERACTION IS BROKEN
AND WE LOST THE RESISTANCE TO
THE DRUG.
SO WE DECIDED TO TRY THIS MODEL
AND SEE WHETHER BY STARRING WITH
THIS ASSUMPTION WHETHER WE CAN
FIND SOME INTERESTING
INTERACTIONS THAT CAN BE
JUSTIFIED.
SO IN GENERAL WE ARE LOOKING FOR
THIS PART OF -- ECONOMIC
LOCATION, SO THAT WHENEVER THEY
BOTH ZERO OR BOTH ONE THEY USE
THE RESISTANCE OF DRUG BUT WHEN
ONE IS ZERO AND ONE IS ONE WE
HAVE ANOTHER STRAIN.
IN CASES WHEN YOU HAVE -- IT'S
EASIER TO SAY THE PROBLEM THAN
TO SOLVE IT BECAUSE IT WAS IN
THE ANALYSIS IF WE TRY ALL THE
POSSIBLE PAIRS, WE ARE LOSING
STATISTICAL POWER BECAUSE WE
DON'T HAVE THE DATA TO
COMPENSATE FOR LOSS OF POWER TO
THE MULTIPLE HYPOTHESIS TESTING.
THEREFORE T FIRST STEP IN THIS
APPROACH WAS REALLY TO USE IN
THIS CASE A GRAPHICAL APPROACH
TO SOMEHOW ZOOM TO FILTER THE
PAIRS MORE PROMISING SO WE WILL
ONLY TEST A SMALL NUMBER OF
THEM.
SO WE APPLY THIS TO EASE DNA
REPAIR PHENOTYPE, THIS IS DATA
AVAILABLE THROUGH ONLINE FROM
THE STUDIES DONE IN 2008.
WE LOOK AT THE RESPONSE TO NPO
TREATMENT BECAUSE IN THIS
PARTICULAR CASE WE HAVE THE
SITUATION THAT IS CONSISTENT
WITH OUR MODEL, NAMELY BOTH
PARTS WERE RESISTANT.
HOW ABOUT PROGENIES WE HAVE 31
SENSITIVE AND 53 RESISTANT.
APPLYING OUR MAP WE IDENTIFY 5
ENTERACTING.
AND INTERESTINGLY THE FIVE WERE
ALWAYS ENTERACTING WITH L-2.
L-1 WITH L-3 AND SO ON.
SO INTERACTING WITH FOUR, FIVE
OTHER LOCI SHOWN IN THIS
PICTURE.
SO WHEN DOING THIS COMPUTATION
WE GOT WORD THAT TRAY IDEKER AND
HIS GROUP DOING INTERACTION
EXPERIMENTALLY IN THE CONTEXT OF
AGAIN DNA DEMON STRAY PHENOTYPE.
THEN YOU USE EXACTLY THE SAME
CHEMICALS AS WE DID.
BUT WE THOUGHT THAT IF -- I
SKIPPED ONE.
THANKS.
WE LOOK AT THE LOCK WITH L-1
MANUAL TO SEE WHAT'S IN IT.
AND LOOKING BY GENE THERE WAS
ONE GENE RAD 5 WHICH ATTACKS
THAT INVOLVES IN THE DOUBLE
STRAND BREAK REPAIR PROCESS.
WE HYPOTHESIZE THIS GENE IS
ACTUALLY HALF OF THOSE EPISTATIC
INTERACTIONS SO NOW TO TEST THE
THE HYPOTHESIS WE ASK TRAI TO
LOOK AT HIS DATA AND IN HIS
PHENOTYPE WHEN THEY GO PAIR BY
PAIR, THIS RAD 5 OCCURS AS A
HAPPEN AND IN FACT IF YOU CAN
CONFIRM THAT THERE'S ALSO SOME
OTHER GENES THAT HE DISCOVERED
IN THIS LOCK THAT WILL INTERACT
WITH RAD 5 AND SO ON AND HE
CONFIRMED THOSE PREDICTIONS SO
THAT SHOWS THAT THIS VERY
ABSTRACT WAY OF LOOKING AT THE
THE INTERACTION, SIMPLE
APPLICATION, TO ZOOM IN AND
IDENTIFY DRUG RESISTANT GENE.
THIS IS INTERESTING BECAUSE THE
KIND OF THING YOU CAN TAKE ANY
PARASITE THAT IS INDEPENDENTLY
REQUIRE -- ACQUIRED DRUG
RESISTANCE OR CAN EVEN FORCE IT
IN TO LAB AND GROW IT UNTIL IT
GETS RESISTANT DRUG THEN CROSS
IT AND YOU CAN APPLY THIS MATTER
TO PINPOINT WHAT ARE THE GENES
IMPORTANT FOR DRUG RESISTANCE,
THIS IS REALLY PERFORM TO KNOW.
WITH THAT, I'M OUT OF TIME SO
TIME FOR ME TO CONCLUDE.
I HOPE I CONVINCE YOU THAT WITH
THOSE VARIOUS LEVEL APPROACHES
STEP OF APPROACHES USING
COMPUTATIONAL INSIGHT TO SUPPORT
THE EXPERIMENTAL DATA WE CAN
START BRIDGING THE GAP FROM
GENOTYPE TO PHENOTYPE.
I SHOW YOU'RE WORK ON HOW
VARIATION IN DNA AND RNA
CONFIRMATION MAY AFFECT
EXPRESSION OF THE GENE.
HOW THOSE VARIATION IN GENE
EXPRESSION WITH PROPAGATE
THROUGH THE NETWORK DISREGULATED
THIS WHOT PATHWAY AND HOW TO USE
COMPUTATIONAL METHOD TO PROPOSE
PATHWAYS DISREGULATED IN THE
DISEASES.
HYPOTHESIS NEEDS TO BE TESTED IN
THE LAB TO BE CONFIRMED.
ALSO THAT HOW THOSE
COMPUTATIONAL METHODS CAN
SUGGEST AN EXPERIMENTAL
TECHNIQUE TO IDENTIFY GENES
INVOLVED IN DRUG RESISTANCE.
OUR COLLABORATORS AS I WAS -- I
COULDN'T DO IT WITHOUT WONDERFUL
FELLOWS IN MY GROUP, THE PEOPLE
AND WONDERFUL COLLABORATORS.
ALSO AS YOU CAN SEE THEY HAVE
DIFFERENCE EXPERTISE.
SO (INDISCERNIBLE) IN
COLLABORATION YANG HUANG DID THE
EPISTASIS AND GRAPH THEORY,
YRKOO-AH KIM DID THE CANCER
NETWORKS OPTIMIZATION.
RAHELEH SALARI AND DAMIAN
WOJTOWICZ.
AND I'M VERY GRATEFUL TO THEM
AND TO YOU FOR HAVING ME.
THANK YOU.
[APPLAUSE]
>> I'M SURE TERRY WOULD BE HAPPY
TO ENTERTAIN QUESTIONS.
IF YOU HAVE THEM TRY TO GET TO
THE MICROPHONE BECAUSE WE ARE
BROADCASTING THIS.
CHUCK.
>> THANK YOU VERY MUCH.
A WONDERFUL TALK.
WE HAVE A QUESTION FOR YOU GOING
BACK TO YOUR DROSOPHILA DELETION
STUDIES AND TRYING TO GET A FEEL
FOR REDUCTIONIST GENETICS HAT
HOW PRACTICAL IT IS TO ACTUALLY
IDENTIFY THE MASTER OR CAUSAL
GENES IN THAT AS OPPOSED
BASICALLY PULLING THINGS OUT OF
POTENTIAL BACKGROUND NOISE.
AS EXAN PM PL, ONE COULD
ENVISION INSTEAD OF LOOKING AT
WHOLE ADULT ROUND UP SAY YOU
SEPARATED THE HEAD FROM THE
REST, YOU MIGHT SEE DIFFERENT
PATTERNS OR IF YOU STAGE THEM
HOW OLD THEY ARE, YOU MIGHT SEE
PRETTY PATTERNS AN EVEN THAT
IT'S CRUDE PATTERNS.
HOW MUCH NOISE IN THERE?
>> THE MASTER REGULATION,
REGULARLYTOR AND WE LOOK AT THE
CANCER DATA WHICH ARE IN ONE
CANCER TISSUE, YOU'RE COMPLETELY
RIGHT.
THIS IS WHERE WE'RE HAVING --
WE'RE TRYING TO LOOK AT THE
INDIVIDUAL TISSUE AND SEE HOW
THE NETWORK AND EVEN HOW THE
NETWORK WILL BE CONSTRUCTED
BASED ON EXPRESSION CORRELATION
DIFFERS BETWEEN THESE TISSUES.
THAT IS THE FIRST QUESTION.
OF COURSE THE FACT THAT WE HAVE
ALL THE DISEASES TOGETHER MUDDY
IT IS PICTURE A LITTLE BIT.
>> I THOUGHT THEY WANTED TO
ANSWER IT.
>> SO SEVERAL TIMES YOU MENTION
THE FACT THAT YOU HAVE EITHER A
LIMITED AMOUNT OF DATA IN TERMS
OF SAMPLES OR TOO MUCH DATA IN
TERMS OF INTERACTIONS.
WONDERING WHAT YOUR THOUGHTS ARE
ON SCALING EXPERIMENTS LIKE GWAS
HOW MANY PATIENTS DO YOU NEED
AND HOW DOES THAT RELATE TO HOW
MANY DIFFERENT POSSIBLE
INTERACTIONS THERE ARE IN THESE
PATHWAYS.
SEEMS TO ME LIKE A LOT OF PEOPLE
ARE GOING AND DOING THESE
EXPERIMENTS WITHOUT REALLY
THINKING A LOT ABOUT HOW THEY'LL
ANALYZE THEM LATER.
>> THE BIGGEST PROBLEM, THE FACT
THAT MOST OF OF THE PHENOTYPES
THAT WE WOULD LIKE ARE COMPLEX.
WE DON'T KNOW IN ADVANCE HOW
COMPLEX IT IS.
SO IN SIMPLE PHENOTYPES YOU HAVE
ONE TO ONE RELATIONSHIP BETWEEN
GENETIC VARIATION AND IF YOU ARE
LUCKY, THAT YOUR PHENOTYPE, WHEN
YOU CAN GET AROUND WITH WHATEVER
OF CASES.
IF PHENOTYPE IS COMPLEX, I DOUBT
ASSOCIATION STUDY WILL ALL BY
ITSELF, YOU HAVE TO HAVE SAMPLES
AND THAT'S WHY WE'RE LOOKING AT
THIS NETWORK LEVEL BECAUSE IT
PROVIDES ADDITIONAL OR
ORTHOGONAL JUSTIFICATION WHY
THOSE TWO ARE RELATED.
BECAUSE NOT ONLY YOU FIND
ASSOCIATION, WE HAVE SOME
COULDN'T OFF HERE.
THIS COUNT OFF, IF YOU TAKE A
STUDY WOULD BE THAT.
THIS IS TOO LITTLE.
SO THOSE ARE POSSIBLE
INTERACTIONS BUT IT WILL NOT
SURFACE MOST OF THEM THAT WE
FOLLOW WITH THE NETWORK APPROACH
WHICH OF THOSE WE CAN SEE
SIGNIFICANT CURRENT GOING
BETWEEN HERE AND THERE.
THEN YOU CAN COMPUTE THE FUTURE
FOR THIS.
THE FUTURE IS FOR THE GWAS STUDY
THAT WE DO, NETWORK LEVEL
ADDITIONAL EVIDENCE TO SOMEHOW
REDUCE THE IMPACT OF THOSE
PHENOTYPE COMPLEX.
>> I HAVE SET OF GENERAL
QUESTIONS, TERRY.
IN CONSIDERING THE COMPLEXITY OF
BIOLOGICAL SYSTEMS, IS IT
CONCEIVABLE THAT THE
MATHEMATICAL TOOLS THAT WE
CURRENTLY HAVE ARE NOT -- REALLY
JUST AN APPROXIMATION TO WHAT WE
WERE EVENTUALLY NEED TO
UNDERSTAND SOME OF THESE
NETWORKS.
ARE THERE ADDITIONAL MATHEMATICS
THAT NEED TO BE DEVELOPED THAT
COULD HELP IN BIOLOGISTS ANALYZE
COMPLICATED SYSTEMS?
>> 
[LAUGHTER]
>> I THINK THIS IS THEY HAVE TO
DEVELOP AS THE BIOLOGY INSIDE IS
RISING SO DEPENDING ON THE DATA
AVAILABLE.
OF COURSE WE HAVE QUITE THE TOOL
OF MATHEMATICAL TOOLS WE CAN USE
MATHEMATICAL EQUATIONS TO
IMAGINE THINGS PRECISELY BUT WE
DONE HAVE DATA TO DO THAT.
SO IT WILL HAVE TO BE DEVELOPED
ALL THE TIME BECAUSE WE'RE
SURPRISED BY THE TYPE OF DATA WE
HAVE AND IN ORDER TO MINE THE
DATA, IN ORDER TO UTILIZE IN THE
BEST POSSIBLE WAY WE HAVE TO
LOOK BACK AND SEE WHAT WAS THE
MATHEMATICS.
IN AN IDEAL WORLD WE HAVE
NETWORKS WHICH ARE PRECISELY
DEFINED WHICH HAVE ALL THE
EQUILIBRIUMS AND ALL THE
PARAMETERS THAT -- AND WILL HAVE
COMPUTERS -- I DON'T THINK THIS
IS GOING TO HAPPEN SOON.
>> IF THERE ARE NO OTHER
QUESTIONS, LET'S THANK TERRY FOR
A VERY STIMULATING TALK.
[APPLAUSE]