Well.

That's what.

I'll.

Welcome hundred zero zero zero S.
professor of biomedical mikes of

Columbia University the division
of that biomedical thinks he

is one of those people who
spearheaded if not for.

Record.

Yelling out you know.

The one that just discovered
you gotta walk so

congratulations it's always
good news on the very.

Last minute ten seconds to
the banks the rapper and

he's also the resources director
of White America the government.

I'm just capable I'm
sure you said the truth

as to why all of that research
is fundamentally what I just.

You know why birth right this is.

A really hard core this
study chaotic behavior

in high dimensional dynamics is
very appropriate for an Italian.

Guy out of chaos that's something
I'm going to try to make it all work

of organized chaos of what
was the disorganized chaos.

Here in Spain at the University of orange
I learned yesterday which I realize.

More observers you know folks I did not
realize that the cities were pushing for

a great fight.

You know.

You spend dollars I shouldn't have been
right sorry if the post look like you

were even older than giving
in the ranks of the manager.

Over there but.

Then he went off into industry for
a while where he was the founder C.T.O.

an executive I present you know.

And that was three point one person so
here's an example.

Wilder's.

Biological systems and
more from the ape or.

Thinker so.

Great matter and yes very much.

And he was hungry was
one of the first people

at the foresight to try to reach what.

He is on the border with
by the National Cancer and

there's a number of scientific
advisory boards and percent of.

All as I said he really is
the only major visionary

it's just my home.

Base and it's a real.

Well to talk with about
a very important problem.

Which is trying to understand power or
normal cells or cancer cells on and

off are and he's got a very novel
approach very difficult problems

associated with brain injury talking
about taking mastery of only

one of the switches
extremely unpredictable put.

Thank you Jeff.

So.

So I actually I have to say that my
relationship with the Georgia Tech started

about fifteen years ago when I heard
Bob Merriam at the National Academy of

Engineering say that compositional biology
was really at the crossroad of three

disciplines was chemistry biology and
computer science and

he said you know what to say about those
if you move his biology if you think

his chemistry and if you failed his
computer science so if it stinks and

fails to move it has got to be
competition on biology so I.

Actually had a slide with that for many
years in the now don't use it anymore but

I thought he was it was very interesting
that he had that vision of integrated.

Integrative science so
this is a cartoon of how

our approach to understanding
biology not just in cancer but

across the spectrum of both basic
biological know physiologic phenotypes and

disease rate if you know types of change
over the last ten years at some point we

thought that one day we'll have
the power to sequence everything and

given enough a large
population either of cells or

of individuals will be able to just look
at the association between traits and

their underlying genetic components and
this turned out to be spectacularly not

the case you probably have seen some of
the papers that were done very deep you

know typing and sequencing four or
six different complex diseases and

basically almost nothing has come out of
the other end and right now the reason for

that is that not only the entire
set of sort of the pet independent

variables that determine if you know
type is much broader because it includes

a very large at the genetic component
that is now starting to be studied but

in fact that the dynamics of the cell
is a major element that drives.

And point phenotypes in way without
even having to change the genetics and

get it Jandek layers and so it becomes
the name of the game I think and

to a large number of people in this
audience should be no surprise is this to

some extent change from studying and
you know sort of built building

zoology of the genetic epigenetic and
functional layers in cells and individuals

to try to understand how the regulatory
logic of the cell manipulates and

integrates that information to produce
a particular trait and disease so we call

these building the assembly manual of
the cell you know if you're trying to fix

a car probably the first thing you should
do and what the first thing the problem

most people do is to actually buy a manual
that tells you how the car circuitry

works trying to fix a car Q trying to
get a manual of assembled a clock.

We don't have an assembly money for

the cell and so about ten years
ago we took up this challenge and

try to see whether we could actually
come up with a more reasonable set of.

Regulatory processes that
integrate that information

why is that important let me give
you a very simple example this is

a cartoon thirty thousand foot view of why
it is important but I watch you try I will

make these extremely tangible in the in
the remaining in the rest of the talk and

the idea is that I repeat these
slides about a thousand times and

you'll have a sense of what today a genome
wide association discovery study origin

expression differential analysis studies
looks like you have a bunch of genes

some of them turn red because they are
either overexpressed or under expressed or

maybe because they carry a particular
genetic variant that seems to be

associated with it with a phenotype you
have absolutely no way to from this mass

trying to to identify which genes may
be more relevant for the disease and

in fact because of limitations on
our ability to identify genetic

operations and and differential expression
a lot of genes are expressed in the in

the gray area where you really can't
tell whether the difference expressed or

not genes that are green may turn out
to be in fact those causally related

to the presentation of phenotype and

completely missing from this picture
if you now have a regulatory network

superimposed to that view things start
becoming a lot clearer and the reason they

become a lot clearer because now you can
start asking questions of the type given

this structure of the regulatory network
and given that I see this pattern about

aeration What are the gene what is
the gene or genes most likely to

have contributed to the particular
pattern and in this case the surprising

result may be that a gene that in fact is
not itself differentially expressed or

or harboring a genetic variant may in fact
be related to these in fact I'll show

you precisely examples in which genes that
do not harbor genetic variants and I.

Are not differentially expressed will
be identified as master regulators in

specific field times.

OK so this sounds fantastic in theory So

how do we get down to actually getting
it getting into into practice so

we could actually do anything akin to
what I showed in the previous slides

we first had to come up with a reasonable
mark of regular interaction in the cell so

over the last ten years we spent
significant amount of time to put together

a variety of tools and
algorithms that would allow us to dissect

transcription networks post translational
networks post unscriptural networks and

then once you have this map and it
literally took about eight years of time

to start getting out how we can now
interrogate these maps to understand

things like for instance the mechanism of
action a particular drug or the genes that

are associated causally to with the
etiology of a particular disease or with

the particular feature to be transitional
for instance a developmental process and

this is actually sort of a good slide for
you in terms of references

because this was the Agricola Rockne that
I'll talk a little bit about was literally

the first one that was used to dissect you
know why regulatory lab work in humans so

this one has been the first one used for
postals missional networks and

these are a number of tools of we've
created more recently in the macro in

a regulatory network and in fact to
discover new Newmark are an ace and

so I think what distinguishes to some
extent our lab is that we have taken sort

of a approach that is skipped over some of
the simpler organised like equalizer or

yeast and gone and tried to go
directly into a human context which

is provide new which to the extent
the take home message was that it

looked very complicated but actually
some of these methods work much better

human cells than they work
in eastern Ukraine and

it's very surprising result probably
due to the nature of the regulatory and

the redundancy of the networks
in which is higher than used.

The Tropic nature of regular.

So I'm going to go and give you a little
bit of an example of how these approaches

work and with a lot of emphasis on how
these approaches are biochemical he and

functionally validated because there
are literally hundreds of methods and

each one promising to work
better than the other one

the proof of the putting is in the eating
and so to a large extent we have to

experimentally validate everything that we
produce and in fact if you don't have in

your mind a way to think about validation
before you even start putting together

a computational tool you should probably
think about that first and not now work on

the competition too until you have it
your ideas are very clear and the map.

So how does the Iraqi algorithm work.

Motif of.

The lab has been the use of information
theory in fact one of the people

that was one of the major contributor
to some of these algorithm the lab and

man he's now a professor here at Emory and

really using information theory has been
an extremely valuable perspective for

us in terms of the second derogatory logic
of the cell because it allows us to apply

a certain theorem that intuitively very
very simple very very intuitive and

that actually allow it to.

Eliminate an enormous number of potential
force positives from your networks so

this is an example this theorem is
called the Data Processing equality and

it tells you that the information measure
as the mutual information which is in bits

that is exchanged between individual on
a phone conversation is always larger

on the direct path so between people that
talk directly than between people to talk

indirectly so there will be if you measure
information between what Mary knows and

what Jane knows there will be
some useful information but

is the result of this transfer information
which is interacting as a result it will

be smaller than both information
exchange between Mary and Joe and

between information and changing joint So
this is very important because in

a biochemical network one
of the major problems.

Previous approaches cool
relevance networks was that we

associated the idea of course expression
we did of regulation which is in fact

cannot be further away from the truth and
you can see it in this case where you

have a transcription factor regulating
another one which is regulating the target

in court suppression there were you would
have inferred a direct interaction with

one of these targets which would
have sort of misled you and

follow up experimental results in this
case if you can measure of pain for

instance by using a large set of
mockery suppression profiles or

any kind of molecular profile data the
information transfer between all three of

these pairs of valuables so if you
want to have two to have to target and

if one to the target you will find that
the information transfer from T.F.

want to target is much smaller than
information transfer between two fun

to have to have two in the target and
therefore you can eliminate OK.

So this was applied again in two thousand
and four in this paper in Nature Genetics

to try and dissect the regulatory logic
of a human be cell both normal and

tumor related and what we show here
is I like to call these things

ridiculous ohms or fuzzy grounds
because it's almost impossible to

make any sense of what they mean by
the very pretty and what this is showing

here is actually did make proton could
gene which is a transcription factor and

in the large circle the direct
agates inferred by Iraq any

more than that a regulated by make in
a smaller circle the indirect ones for

things that appear to be regulated
through some intermediary process and

what we were pleased to find at the time
was that about half of the ones that we

had detected the one in red and pink were
in fact unknown in the literature and

been and had already been experimentally
validated so we said that's great there is

enrichment but as I said this is no
longer there is no longer allows you

to get away with any kind of biological
conclusion and so we took twelve

of the white circles that had never been
previously reported make targets and

we do it something called Press.

With essentially measures whether
the MacPro thing binds up stream in

the promoter region of the corresponding
targets and as you can see here this is

the this is the negative control this
is a positive control and is the result

showing that in fact for eleven of the
twelve genes that we tested at the time

we could show very definitely that Mick
was biting the promoter region and

this is been largely improved there are
several papers in informatics that really

expands on the theoretical component of
the now of the analyses and also in nature

protocols where we introduce
bootstrapping in different type of.

Informational estimators And now from the
exact same data which was a collection of

about three hundred fifty marker is
pression from file from human the cells

we can identify about three hundred or
more targets of MC and in fact with a much

higher even higher today and
what we originally got validation rate.

OK So let me tell you this
is great that you say OK so

these protein biopsy move make does not
mean very much probably not make a stream

of a lot of different proteins at the time
this was consider about chemical the.

Good proof that the that there was
regulation much description factor but

in practice we no longer use that type of
approach and what we do is we first of

all don't validate just a few targets but
a genome wide all the targets for

the binding affinity to the protein using
something called sheep seek which in your

position bought by sequencing the entire
set of promoters in the human genome and

the second thing that we do is we
functionally validate the targets by

silencing the transcription factor and
seeing whether the predicted target of

the receiver factor in fact affected you
know privately by by that science and so

this is an example of a network that
we used to study pluripotency and

lineage specific differentiation
using a tumor model which is out of

germ cell tumors growing carcinoma
in particular so you know mio.

Carcinoma of course you know most and the
broadcast you know my particular in that

group completely mimic
the developmental paths of a embryonic

stem cell so they when exposed to
Otto B.M.P. they follow exactly the same

type of developmental lineages and so
we did we use a large collection of these

embryonic carcinomas and other germ cell
tumors to build a regulatory network and

what you see in another fuzzy gram here
is just the backbone that history strip

purely transcription factor based So
this the nodes here are only transcription

factor and the edges indicate relationship
between the relationship between

transcription factor and so we asked then
so how can we validate these networks and

we looked at three genes that are
particularly important for to report and

see in fact as you know when you
go express these gene with make in

a fiber glass for instance these different
different shapes into a pool report and

cell that's what the I.P.S.
cells are based on so these are opt for

stocks to a nanogram and
now this is the type of validation

that we do pretty routinely for
these matters it is purely for

giving a sense of how we do this in the
lab today so these graphs are very nice

graph that we use for a lot of different
things called Gene Cernan which Manasseh

simply enrichment analysis plot where on
this axis for instance you're showing

all the genes sorted by the one most
differential expressed following science

you know four to one least differential
expressible in silencing all four and

on this axis you're showing grad
the predicted target or acne and

what you can see mediately is that in fact
the color coding on this part the dark to

light shows the density of the red bar and
so you can see there's dramatic enrichment

in density very close to the targets
that have changed most and

this is repeated for all three of
the transcript in fact a lower salary.

And this curve here is one a curve that
goes up a notch every time you hit a bar

and goes down a notch every
time you hit a non bar

with different weights given to bars and
Omar such that when you start and

you start and
end at the same exact place OK so this.

Developed the Broad Institute
by public Tamayo and.

And removed and the Colgin set a mission.

So you can see here functionally these
inferred targets seem to do exactly

what we predicted that is you know in
a particular law in a particular line when

we silence them the targets that we
predict and move on this side what

you're seeing is again all genes
sorted by this time they're sorted

by the probability of their promoter
binding the jurisdiction factor so

these is the one with the highest
probability of binding out for this

one with the lowest probably abiding order
for according to chip seek analysis and

again you see this case is complete in
this respect activity if you look at

the enrichment there's almost like a total
black bar there at the top which means

that the density of the red bar is really
very very high right at the beginning and

in fact we were validated by university
Taishan individual targets ten out of ten

ten out of ten and
nine out of ten in one cell line in ten or

twenty other cell line validated for the
same gene so this gives you a sense of how

accurate the prediction of our acne are
they typically range in the seventy two so

you're ninety percent in back and follow
up by chemical ideation and do they work

for a prescription factor they don't so in
fact we published a paper a notch one and

we showed the north to predict which one
targets we actually have to estimate not

one protein expression actually
active peptide as a billion

a nucleus because some of the proteins
are heavily posterization modified and

therefore them are in a really does not
relate to the truck to fuel the protein

but for about seventy percent
of transcription factors

we're able to reconstruct these
now were very accurately.

OK So the question now is
does this work on LI for

reconstruct infrastructural networks or
can we use these approach also to

reconstruct more complex by chemical
networks recent signalling networks or

metabolic pathways and I won't talk
about metabolic pathways actually an M.

in mine and one of my students
ever worked to newspapers just

been accepted to show that Iraq can be
used to reconstruct metabolic pathways.

Using also mass action
law conservation but

this is a particularly interesting
result because it uses.

A set of data that was generated
by the Michael Cohen Group at cell

signal a technology.

That essentially profiled fossil peptides
in non-small lung carcinomas and

the idea to take tissue licet for about
two hundred different tumors samples and

cell lines you know basically take lightly
digesting it with restriction enzymes and

then immune to purify it with antibodies
that detect forcefully Tyrus and

so this is possible terracing
specific antibodies and

then huge essentially use time
to master trust to simply

sequence these peptides hope to have
seven to twenty I mean acids and

then you get a map of the abundance
of these peptides in the cell and

so you can treat the suspension the same
as an expression profile except this time

you're not measuring expression of genes
you're measuring the availability of

the hospital and there's a lot of changes
you have to do to Iraqi because there's

couple of problems one this data is highly
discreet while Iraqi people who tends to

use continues data the other one is
that there's a lot of the generation

the same protein may contain multiple
fossil peptides and actually the same

fossil peptide in terms of sequence
maybe long to different proteins so

we have to deal with those results but

this is kind of the result
analysis this is E.G.

if I was one of about eighty nine kinases
that were profile but his method and

what you see here is the Iraqi
prediction for the substrate of E.G.F.R.

notably these five substrates in red and

yellow were the ones that were known in
the database in the in the fossil site

database us as targets of E.G.F.R. so you
can see that the prediction is a little

bit larger than what we actually
knew about E.G. far and.

What you're seeing here in
green our substrate that

we're validated by Psylocke us is the
measure essentially the change you fall.

Relation of the substrate bowing
silencing of the of the kinase and

so you're seeing that at this stage with
the least restrictive Iraq needs threshold

about sixty two percent of the targets we
were identifying by the method by day by

Psylocke if you use a slightly more
restrictive threshold from Iraq any you

now get into the ninety percent range and
still have been depicted in yellow or

red but they're still all here
they're known targets of E.G.F.R. So

this allows us to build a full
compliment of tire seen kinase

phosphorylation substrates which
now it generates a network for

first initial process that's
typically been quite hard to get.

And I'm going to do something now that I
always discourage people from doing which

is illustrate the result of something to
something hasn't been introduced yet but

since I'm going to do it right away and
this was better explained this context

to see this and so we use this
algorithm called master regulator and

I see that I will stray later
to now take these phosphor.

Fossil signalling networks and interrogate
them to identify what are the potential

master regulators of the specifics up
types of the tumor in each cell line

that was available in our NASA any tumor
sample So first of this is a particular

cell line in which the records and
Michael come group that found out

fusion construct as being causally related
to two more Janice's there is another

one in which Matt was identified as
the causal lesion and is one one for P.G.

a far and when we were on master regulator
and I think on this network as you can see

these are the master regulators that we
identified and as you can see from these

two lines are can met were identified in
this line we did not and far because in

fact if you did you fart component of the
network is not well profiled essentially

from the force were the only data we
don't get enough data for pretty far for

one reason or another and so we can't
represent the well in the network but

what was remarkable is that is well
known doubt silencing are the out.

A fusion Gene has no effect on
the viability of the cell and so

this is a triggering.

Sort of.

Transformation but it's not necessary one
of the tumors addicted to however when we

call abrasion which is the Southwestern
Michael White and John Mina we went and

tested the other regulators there were
ninety five we found that in fact we were

discussing efforts to hear we
found the P.K. two and you know F.

or two in isolation but not in
combination seem to be actually quite.

Due to have seen it in effect and
so by ability so

this is an example and we're now doing
it for every single cell line and

we're doing all the possible combinations
of these to identify mass regular of long

term Regenesis and what is interesting
here is that we were able to do this at

the single cell level meaning that instead
of identifying the mass regulator for

a subtype of the tumor we could tell
Well these are the protein does seem

to be the tumor seems to be addicted
in this particular cell and

this particular tumor and so
this is kind of an opening for

what we're now trying to do in terms of
more personalized approaches to their.

OK So everything I've told you
is kind of a big white lie and

white while I that entire field keeps
repeating which is that makes it look

like everything in the cell is regulated
pairwise fashion OK almost nothing in

a cell is regular impair wise fashion and
so this network representation that we see

where you see two nodes connected by
an edge really don't mean anything because

that edge may or may not be there the
pending on a variety of other proteins so

we try to see whether we can still use
information theory to dissect these

modulation up effect where for instance
the ability of a transcription factor

to regulate its targets is modulator
both post transcription of meters

on the protein or poor or it genetically
has on the target gene by a variety of

different proteins including protein diet
due possibly some modification or actress

relation as we've seen of the target
prescription factor or co-factors etc.

So the idea was again from information
theory that if you have a conversation

between two individuals and
it's not going that well but

there's somebody at the phone company and
somebody Tony starts moving around some

dials at some point the conversation may
improve and when that improves we this is

captured by something called a conditional
mission information what we demonstrated

essentially theory and that shows that
the conditional information is the optimal

task to detect these type of relationship
three three Vajra relationship where you

have this kind of modulatory effect and
this app is exactly biochemical networks

again because for instance if you look
at MC make it can be transcribed and

translated into the meat protein but
if there's a signal coming on the just

a three cascade make gets this POS
related story in sixty two and

tagged by you because the nation and
therefore very rapidly eliminated

down degraded by the by the Proteas
ome and as a result even though

you may have a lot of them are unable
make you have very little protein and so

the correlation between the money of
MC in the American way of a target of

make is pretty bad exactly like the
conversation if you now have a blade that

pathway what happens is dynamic
is no longer degraded and

the correlation between
money of making the M.R.

and A of the target becomes very high so
again we devised this was published last

year in nature about ecology and arguing
coalmine the that was able to essentially

interrogate genome wide all
the possible transcription factor and

all the possible modulators to identify
which protein seems to modulate which

transcription factors and the way it works
is actually quite easy to understand so

imagine that each one of these
columns in these two hit maps

is a expression profile.

These two blocks represent the set of
experiments in which you have the thirty

five percent highest and thirty five
percent lowest expression of the modulator

you're interested in so in this case
searing through an ace experience or

any kind is thirty eight and then what
you can do is sort these again according

to the expression of the transcription
factor of interest in this case make so

little make I make will make I make and
what you're seeing here.

It's really a very interesting
pattern that shows that

on this axis you have the targets you make
in the starkest are not regulated by make

when there's low levels to get thirty
eight They're beautiful the repressed and

activated when there's a high
level of us together here so

the assumption here is that you see
something like this maybe Ezekiel thirty

eight seems to be a potential modulator
of make up till we found out in fact

educated is a very potent modulator
make activity it binds make and

he actually binds make differentially
either in the center domain or

in the entering of domain
depending on its activity so

if it's postulated it doesn't bind the
central domain only binds the interim of

the main If it's not force related so we
knocked there it by the Central to me when

he bind central to me and he protects me
from postulation a certain sixty two and

so make it super stable you get
super mic it never gets degraded but

when it's on the entering of the main it
actually force relays make a certain sixty

two possibly through to mediate
a false relation signals and

as a result this is what happens that if
you actually silence S.T.K. thirty eight

or actually if you ever express it which
is another paper that is now review.

The rest of your career is gone and

meat protein is completely gone of the
protein level but unchanged in a lab and

so this can be done in
fact there were other

papers where we actually in this case used
both Mindi to identify a modulator of and

make inroads cells and actually defect or
three a complete abrogate the effect of

the of these protocols to be conjugated
called you we want and again you

can see that when you want a science make
a level go through the roof for the R.N.A.

level and change and unfortunately there's
a lot of light here you can see it but

essentially if you lose you want in
the brain you now start having major

morphogenesis the facts and
with accumulation of a second you want but

if you're a D.L.L. tree you
completely rescue the phenotype and

all you find the modulator but you can
find through Iraq near the actual effect.

OK so here's another one of these.

This one is an interesting one was
published over a couple years ago and

what it shows is the first map where you
can actually associate signaling and

drug targets to transcription factor
regulations so everything in yellow

is a transcription factor everything in
purple is signalling protein in particular

signalling proteins that are known to be
drug of all and what you see in some of

them like the able kinase are both
a transcription factor and.

And what you're seeing here doesn't make
any sense and you cannot interpret it but

is in fact completely machine
interpretable and provides you with what

we call a drug ability matrix from the
cell which to you basically what you want

to say is two things if I want to activate
or repress a particular transcription

program this may just will tell you
how to get it done using drugs and

if you want to actually know which kinase
had contributed to a particular phenotype

this can be also interrogated to find
out and I'll show you some results.

So I showed you some interesting results
were proteins the module a transcription

factor activity but obviously you can
do this for almost any model to affect

research and you could use it to
identify whether a particular snip

changes the ability of a protein to
regulate sagas you could use it for

you know understanding whether
there are sort of third order or

fourth order regulatory processes and this
particular case is what we tried to do is

to understand micro and
a biogenesis and the idea is that

came up completely serendipitously
from the fact that we could not for

all the possible outcomes that we made
express mature mirror in a burka cell and

the precursor was perfectly expressive
of the true could not form and

when you put the same construct in
a different so it was just working fine so

we thought there must be something that
participates to nearby Janice there is

still a specific to this mirror because
the other near are working fine in

inverted and that is all too
specific to a cellular context this

is a kind of a little map of the back
Genesis process for a black or an AIDS and

you know dice.

And and the risk complex accent except or

are very well known this about
fifteen to twenty proteins that

are known to participate but they're
all really not complex specific and

they're all non really macro in a specific
and well we sense seem to find is that

there are a lot of things that seem
to be totally maccaroni specific so

we use this approach where in this case
we could estimate the mere precursor by

taking mirrors that are contained within
host genes so that we could use the M.R.A.

of the host gene like for us and the locus
host near fifteen and near sixteen so

we could use their money of the gene as a
proxy for the expression of the precursor

of the macaroni and that we could measure
the mature mirror using expression arrays

and so what you're seeing here
is very interesting when D.D.X.

ten is low the mere precursor estimated
through the proxy for mirror to eighteen

rank from low to high does not clearly
did all with the mature Mir But

if D.D.X. Tanny's high you can see
an almost perfect correlation and

these are not spectacular now correlated
from completion correlate and

this shows an experimental validation
factor when you silence the X. ten you end

up getting a significant difference in
the availability of the part you are me.

Is another example where you see exactly
the same situation if you look at before

in fact there is an inverse correlation
between the ability of the person ability

of the of the mature but
if you now express our B.P.

for the highest level you now get
a very good positive correlation and

again we validated this by experimental
osses So if this is true for

men by Genesis you can now start
thinking about using Also these for

mere activity so for instance there was
a very nice paper that appeared nature

a couple of months ago but the polyp and
all three that showed that

the P ten pseudogene which is
completely inactive has absolutely

no function that is known of X. is
a decoys for a particular Mere that size.

And says pretend as a result when there's
a high expression of the P ten gene

it acts as a dick or intact or
it's a way the back or an A and P.

ten is no longer suppress But
when the pseudogene is down regulated

then pretend is very high suppressed and
so the pseudogenes is in no effect

it is an awkward gene because when
it's area activated it suppresses B.

time which is the most
impressive kind of nice which.

So we ask the say the same
question we say well if we now can

profile the mature mirror and
we can profile it's.

Are there genes the modulator the ability
of the mirror to regulate the target and

again we don't if I both regulators
are seems to be agnostic

to the mere types of regulate the activity
of many mirrors and we identify.

Modulators that are in fact very specific
and this includes both genes that seem to

be acting as decoys as well as genes
that are in a bind in genes and may

modulate them you're using other mechanism
this is experiment of validation for

three of them including me one of
six that essentially buying P ten so

this may be forty cents another decore for
the near one of six which was the one

identified one hundred actually
we think was near twenty one.

OK so once we have these completely
complex specific information

on regulation we can now start
bringing in additional information and

we are very very cautious about this
because most networks the people of you so

far have been basic like protein
protein interaction there are bays and

used to hybrid Ses which are not
context specific and in fact are X.

we were asses which if you which
are very unlikely to predict

the ability of the two proteins to bind
in vivo at physiologic conditions this

protein as way of expressing that he has
to host they are in an infant can combine

offers are the condition they are not
forced for laser because terracing kinases

are not present in the east and so
the environment is totally different and

if you believe it you know
the regulatory networks are differ.

In each one of the cells and so what
you get X. Bebo is just one incarnation

of all these possible subtypes of networks
so we like to bring in this information we

like to bring it in a second stage and
this is done through evidence integration

I've written essentially as a Beijing
approaches you can use random forest or

you can use the I base it turns out
that it really doesn't make much of

a difference why do you use the advantage
that you get by integrating more than one

modality way overshadows whatever our
competition to use for integration and so

we can do it for protein protein protein
in a now we also have propeller mirror.

R.N.A. interactions and and these are now
starting to be very very specific and

we publish a paper recently Mark our
system biology where we use these

compute integrate approaches both
to identify a novel complex that

combines the the.

The premedication complex
with my code to complex

thought to control complex which
are known to be functionally related but

never shown to be in fact physically
interacting in the protein and also that I

defy the mass regulators of germinal
center formation call me been folks want.

And we're now sort of funded to apply
these type of approaches to variety of

tumors and in fact non tumor
related phenotypes particular

area last injure germ cell tumors and
number of other stem cell.

Initiatives that were recently found but

these are in the cell
breast cancer prostate.

And these are for D.C. J.
complex and non-small cell lung.

All right so so that's great so

we now i hope you believe that
we can do at least a decision.

Effort of trying to reconstruct some
of these regulatory interactions.

Now what OK so how exciting is it to
be able to say these are the targets of

a particular a regulator maybe
interesting about chemical but from about

perspective it tells you not very much in
fact may even make the entire game more

complicated because now you're staring
at these incredible fuzzy grams and.

You don't know how to interpret so
why we wanted to do is to go from you know

an expansion or we say analytical
approach to now it's in that approach

that would allow us to zoom in on a very
very small number of candidate genes

that are causally related to the
presentation of particular phenotype and

we introduce his medical
mass regular analysis

that has been really much more
successful No we expected it to.

But that is also extraordinarily
simple conceptually understand and

the idea this is the following imagine
that you have two phenotypes one is normal

one is tumor related or maybe two
phenotype in a tumor progression cascade

the idea is that if you imagine that there
is one regulator that is responsible for

this transformation one or
more let's not start with one then if your

network is a very accurate representation
of the targets that are repress

are activated by these regulator is it
directly if it's interesting from Fox or

indirectly if you say kinase then it is
very reasonable to assume that if this is

an oncogene Well exactly where the target
should be over expressing this to shew and

its repressed target should be
under expressed in the tissue and

you can actually measure this if if
this representation is correct and

that's what you should see you can
now measured using these kind of and

Richmond analysis approaches so
what this is showing is this

all genes sort of from the most over
Express in one tissue for the most

under expressing the tissue compared
to you or your other field type and

what you see Again you the bars
are the predicted targets of this

restriction factor T.F.T.P.
one and again what you're saying

is that this targets all very nicely
accumulate a one end of the spectrum and

therefore you can associate a certain
probability that these are provided if you

want may be associated with idiology
of these particular disease and.

If you do this for

every single transcription factor quite
a lot will have this type of shape and so

now you have to figure out which one works
best and we did this by using a number of

methods one uses sort of a an odd
risk ratio approach that allows.

Rank the genes if you rank the gene by
people you get horrible results because

this Preval here is complete
dependent on the size of the regular.

And the other method that we use is
a regression methods Well basically allows

us to sort of formulate a pseudo kind
adic of the regulation process so

that we can basically among the genes that
come up US candidates say fifty what is

the rank of the top five and which one
regulates the largest subset of the genes

that change in that in that context so
let's see if this works what we did here

is silenced three different
prescription factor in a human B.

cell this is a working phone and
the reply to be zero six Fox and

one in name and we asked the question
if you eliminate any R.N.A.

for destruction factor can
you infer which one was.

Regulated by the silencing
as a shining and

as you can see we used two different
approaches for the genes that are richer

in ours is really doesn't make much
of a difference in monkeys they all.

In one case one came in number two with
the original gene separate Eunice's

a method that we've developed in-house but
it really doesn't make a difference

essentially always no matter
what you do in the top that and

this is basically where you want to be
because then it's very easy to validate

things does this work post translational
this is much harder case so

we took the protein that turned out to be
the most clear tropic kinase in a B. cell

which is in fact a sticky thirty eight
which were ready had science in data and

so this is setting yourself up for failure
because this prosing regulates more

than one hundred transcription factors so
the response you're going to have is

extraordinary Has your genius and to think
that you are able to after your science to

reconstruct that that was the protein have
been science it's a big leap of a big

leap of faith and in this case we know
basically what the media are going did for

us it allows to pretty quick transcription
factor effect in particular which

faggot's of those transcription factor
should have been affected now you have

a bigger what we call back.

Extended regular on and we could use
the same enrichment I says on these

targets rather than direct are going
to transcription factors or

you see here that educated eight came
up as number five or seven hundred

seventy two proteins and then in fact
the one that came up a number one C.D.C.

five shares more than half of
its regular targets with S.T.K.

thirty eight so this actually gives Also
since I would now following on that tells

you that you can now start using this
information on how targets are shared

to try to reconstruct the topology of
the signalling pathways because if it

is a signal improve things is right on
top of another want to regulate sit

you soon that this one would be slightly
more people trapped in this one and

in fact cover both the targets of
this one and some other targets.

So this seems to be working fine in-vitro
OK And this is a big leap of faith

because now we have to show that he works
well with an actual biological phenotype.

So the story that I'm going to tell you.

For ASA a very big.

Leap Forward and this type of
approach about tickle systems and

was motivated by a paper by a kennel dopy.

M.D. Anderson who discovered about five
years ago that patients affected by

Lupus normal to form a really bad form of
brain tumour the most it is very rare but

is the most frequent form of brain
tumor that's a story malignant where

segregating in three subgroups so these
are patients and these are genes and red

means over express in this patient green
means under expressing the patient now you

see here there was a group of people that
where over expressing proof genes a group

the worst if you will genes in a group
that was expressing proliferated genes and

these close Agapit with the prognostic
signature with a missing

subgroup doing the absolute
worst in terms of their lie.

Expectancy and so a cancer biologist
would have asked the question

what are these genes they would have
looked at very hard at this group and

in this group and say Is there something
that I know about these that reminds me of

something where we can now ask after we
reconstruct the regulatory network is

who regulates the signature because
if you change one single regulators

regulator are you going to have a massive
effect at the level of the genes that

change then all of them are going to be
causally related to the field type and

in fact what you what I'll show you is
that the genes that we discover and

demonstrated were related to
the phenotype are not in this group and

you would never have found them by just
looking at different expression eyes so

how do how do we do this we first used
Iraq need to reconstruct a complete

complete transcription regulatory
network this is the competition network

that was generated after we run marina so
we used his network and

we asked what other transcription factor
there are according to their regulatory

pattern the master regulators
of the missing subtype OK It

turns out that only five or so everything
in Scion here is a missing pregnant

in the signature showed you this five
transmission five of those are factors

seem to operate you Lady signature this
one in blue down regular decision in fact

this is lost in every tumor is a typical
tumor suppressor genes this is fantasy or

computational fantasies this is reality
because we then what invalidated again

all of these prescription facts are both
in terms of the binding activity so

eighty percent of the test that we run in
terms of running each one of the four for

transcription factor again starting last
one is a negative control issue about

eighty percent of them would show that
they were binding up super star gates and

these are the functional silencing so
we silence every one of the transcription

factor and
show that in fact the genes that are most

differential expressed are precisely the
one in his missing in the signature and we

actually both silence them in human cells
with the expression we over expressed them

in your in cells with very low expression
to see that both were consistent.

One thing that was.

Very interesting is that when we looked
at these five prescription factor and

did a little bit of chemistry on
them we found that they were fine so

Iraq is not very good at determining the
direction if both if one is received after

the other one is the target then we know
the relation can only be going this way

but if there's a transcript or a factor
we don't know who regulates whom and

so we're to do some station to find out
who is buying upstream of the other and

came up with this typology which is
very interesting because it's completely

modular centrally almost every
jurisdiction factor seems to be regulating

the others and is completely hierarchical
everything flowed on this way and

which we didn't expect and to Gene C.B.P.
and start three so he has to serve

units the beta and the deficit units that
came out both very high honor analysis

seems to be on top Moreover what was very
you know really remarkable is that when

you look at the enrichment of inner C.B.P.
targets or start three targets

it was high it was very very
significant about six percent but

when you look at the enrichment
of my Second Mile targets

in the intersection of the regular It
was thirty six percent almost half of

the genes that they call regulate
were in fact and sank you know and

this led us to believe that maybe what
we're seeing here is in fact a senior just

a fact so how we went invited Is
this our mooring stems neural stem

cell when you deprive them of my agenda
develop along a neuron linage So that

becomes neurons you can see here when you
quote topically expressed that three in C.

beta but not if you express only one
of them there's a very significant

morphological transformation
the cell becomes very flat and

then I'm going to look like fiberglass and
you deprive my agenda develop along and.

This is a completely barren linage
because there's no way the physiological

the a stem cell will ever go into
a second linage and what you see here is

really blew up our our minds because what
you see here these are major markers for

type and we also did scratch us
a matter of jelling vision ass's and

show that these cells become.

Really invasive and and migratory
when you when you or express both but

here you're seeing these marker
missing close up type and

if you over expressing this about three or
see the data they tend to go up a little

bit a little notch you know twenty
percent thirty percent fifty percent but

if you call express both of them there
is you know anywhere from a ten fold to

one hundred fold increase in their over
expression so now this shows that these

genes are sufficient to induce a missing
trance differentiation of that OK but

are they also necessary in the tumors or
is the tumor addicted to them and

what we show here is these are now cell
lines from my great human glioma So

these are taken directly from patients and
cultured and

what you're seeing again is not very
visible but the mark in red and

green here are again markers of my
second most up that have been acting.

I think exacting and.

Forty And you can see that SAT SAT is that
removes a little bit of them see beta

silencing moves a little bit of them but
when you silence both of them

they're completely gone and again you see
the synergistic type of behavior and so

now the shows of the both necessary and
sufficient this is still not completely

satisfactory because this is again
in vitro data so we went in vivo and

we did a century intracranial
injections of high grade almost cells

in mice and this applies as
you can see from this one in.

These are incredibly tumor Janick they
all die in about one hundred twenty days

if you silence is that three or so you
made out of the combination you can see

that three they do a little bit better
beginning especially all crashed very

rapidly in all die pretty much
in about the same time with C.P.

beta silencing they do significantly
better but repeated the experiment out of

twelve mice in total so this one this is
a result of the first experiment six mice

but out of twelve mice three of them
died and had signs of the of a group of

sperm where you science book out of
twelve only had any evidence of a tumour.

This is a tool you're seeing here this is
not a stoma this is what it was looks like

essentially like a like a coastline
that has been eroded because the cells

are tumor are migrating anybody nearby
to show such a speed that by the time

the surgeon respects it it's already in
the other hemisphere of the brain this is

a completely mobilized tumor is now my
grating are all in a nearby tissue OK And

so this century proved
definitely we die in vivo

these beans completely change the more
genetic potential of these cells

we also want to again human in
a complete different cohorts and

retrospectively check whether this could
be good markers for progress and what we

saw is that in fact one hundred percent of
the patients that had the double positive

so to answer your data when in fact dead
in about one hundred twenty weeks fifty

percent of the patients that had double
negative when in fact still alive so

this seems to be one of those really
helpful markers for to emergency so

you may ask is your network rigged Are you
always going to come up with a mastery

because no matter what question you ask so
this shows that if you actually

use them a second most unusual
proofs in your political signature

you get very different answer and
they make a lot of sense like for

instance all it too is a well
known regulator of neural.

Targets and you can see that is
lost in the game in the rural but

most of the other mass regulator
essentially completely independent.

And the other question you can ask and the
fact that this was asked by the reviewers

of the Nature paper is well how
robust is the No a lot of people

do the same exact experiment to
the French expression rises and

they overlap the French express gene
is like ten percent how good can you

repeat this these analyses if you
tell you put different data spent.

The is that we run it on
different data sets extracted now

can there interrogating in our complete
new panel which another and this show

is the overlap in the three predictions
century perfect almost complete overlap.

These are the master regulators twenty
two out of about forty that we identified

where in fact by all three datasets
about fifty percent and the ones that

are found by at least two out of three
that number grows very significantly so

only about twenty of them twenty six of
them were not found by abuse two of the of

the math and in fact the five that
were that if I had a module all it.

Remains L.

O. O. O. O. O. O. O. O. O. O. O. O. O. O.

O.

OK so now.

You believe this.

That is that one hundred percent
of the missing subtype is

controlled by these two
genes you have to ask.

I do this genes harboring a genetic
alteration or something upstream of them

and it alteration goes see them
remove them it's not going to be.

It changes the name of the game from what
we call genome wide association studies

to something that we now call pathway Y.
association studies so we used and

this is unpublished data we use both
Mindy and Iraq me to go upstream of

the robbery you couldn't identify
kinetic lesions in copper so etc and

epigenetic operation that were
associated with the phenotype and.

For.

This.

Is the.

Most.

Wide you'll.

Hear.

Hope.

All is.

Just around and all.

That.

Look at.

Our.

Very.

Own.

Home.

My.

God.

Do.

You know all the.

Time.

However we.

Are.

All the.

But I'm there Doc says I can control.

All right so now these allows us to
formulate a new theory of cancer and

this is where things get good fun and get.

You know potentially controversial So
if you are a pharmaceutical company

you have two goals one goal is to develop
appropriate therapeutic targets and

the small molecules the modulating the
other goes on to fight a biomarkers that

can be used to stratified the population
for instance if you had you know hurt too

in breast cancer is both valuable
as a biomarkers and as a target and

if you had to run an air septum trial in
all the general breast cancer population

he would fail miserably so the typical
place where you find both by markers and

targets is so for The Back in fact is both
can be fine in terms of the genes that

Aboriginal acculturation but they can
also be fine in terms of genes that have

differential expression by markers what we
are getting is this is exactly the wrong

place to look why because if you
look here you are you're segregating

you're over segregating your disease only
a very small percent of breast cancer

her two implication only a small percent
of lung cancer have E.G. a foreign

place occasion all the small percent
of the know where the entire disease

have been rough implication So
now you're building a drug that works only

ten percent twenty percent
thirty percent of the cases.

Here it's absolutely idiotic to look for

biomarkers because this is like going
to the scene of a plane crash and

saying the biggest broken piece
is the cause of the crash and

the reality is that you know sitting on
the sea ice and you can watch your brother

will review the progress of the Randy
early detection response network and

they had to validate sixteen markers that
were produced using this approach and

nine out of sixteen validated in studies
and full of studies I believe because this

is the wrong way to look for biomarkers
now that they've done about twenty five.

One of them is validated but that gives
you kind of the ratio between how

many things you have to try and
many things are going to validate so

however note the following There's
an almost infinite spectrum of genetic

alterations almost every patient in cancer
will have its own particular combination

of loss of tumour suppressor and gain of
oncogene and there's almost no general

pattern in the maybe some individual
gene that may have a higher or lower but

in general as a pattern is very specific
However if you look at the molecular

phenotype level things seems to fall into
a small number of categories if that is

the case if you believe this well then
there must be some integrity logic and

what we call master integrators that
will integrate these genetic signals and

if you generate signals and produce
a relatively small number of distinct

phenotype and what we're saying and

I hope I've demonstrated are for
that would be an case but I wanted to also

give you a couple of other references
which are and so this is a global sort of

paper in Nature there's another paper
nation we published last year and

it was largely selling for months where we
showed with regard to my father that in

fact he and if half of the profit in
Africa itself is not genetically altered

but it integrates about twenty different
mutation in the B.C.R. pathway and in D.C.

forty pathway that contribute to
the difference between the A.B.C.

subtype and you should be surprised
if you will be selling forma

which are very very different programs
A.B.C. is very aggressive the C.B.S.

treated relatively well
with current there and

following record resistance in T L L So
this basically says this layer here

is a great layer look at it both for
therapeutic targets and for biomarkers.

How much time you have.

I mean it's OK So let me go very quickly
about drug mechanism action because now we

have the targets and the bio markets can
we coupled them with potential to refute

it compounds so
we looked at whether we could actually use

the Connectivity Map which is a map
created by the Broad Institute which is.

Actually perturbed so
lines with particular compounds and

then build the expression profiles
with two different type of

approaches one where we repeat the same
the same probation with the same drug but

in a different so your context in this
case a diffuse large reselling form and

what we can see is that about you know and

I wish fifty percent of the drugs could
be identified in the Connectivity Map but

about fifty percent really could not be
identified by matching the profile so

cell context is very important and
you can actually map unless

you're going to profile every single
context may have some problems the other

thing that we want to know as well is
a more important question which is well so

this is in vitro How are in vivo So if I
give a particular drug to to a mouse and

I now take a sample from his prostate can
I match it to the corresponding drug than

in the Connectivity Map So we do these
very large projects not to do this but

to build it in people interact
to him in prostate cancer

where we took my from different
fifteen different genetically engineer

strains are predisposed to prostate cancer
and cross them with about a panel fifteen

drugs that would produce therefore a lot
of different perturbation but united and

drug related who are D. prostitution and
we now finish quite a number of them

these are all the ones that we've
already in the ready analyzed and

what was very interesting is that and this
is for this project we want to compare

human interact Ohman prostatectomy was
very interesting is out of the drugs that

we test that could be compared with a can
actually map only one which is right for

my seeing the maps in two zero the most
which is also an M two inhibitor

could be identified as sort of very highly
matching in the Connectivity Map So

when you going vivo you're
even more unrelated

to what happens in the interest of
situation so I want to do is see whether

we can actually get a better sense
of the mechanism of action of drugs

in a way that is independent of the
context because you actually didn't find

out the profile which made involve all
sorts of different processing etc But

actually the real mechanism of action
of the drug and we use these other.

And cool idea was published
in two thousand and

eight which basically says if you
have a regulatory network and

you an opportunity with a drug instead
of looking at which genes changes let's

look at which interactions change so
our gene you may completely lose

it because that's a that has hit the post
translation only by drugs or in a bit of

a kindness you're never going to see any
differential expression of the trainees or

as you can see some complex feedback loops
but if you look at the interaction of

the kind there are going to change
quite significant so we did this and

basically what it does you take two
phenotype drug related one hundred one and

you end up seeing which edges in a
particular regular change and we did this

for two drugs that have almost identical
mechanism of action on a schoolmaster

tracks it was called purple trucks it
they had a an enzyme called the other for

that reductive base in the cell and one
is a more potent trucks individual for.

Protein the important thing is the
protractor also seems to have an amazing

resolved in patients that are refractory
to most attractive utilization map of

selling form so I want to see whether
these comparison could actually do seed

a little bit of the of the difference and
what we saw was very interesting not only

the method identified in the Top twenty
is actually the top twenty six or

thirty for each one of the two drugs the
genes that sit in the most is regulated

area of the network identified yet
referred reductive as one of them but

fifty percent of the other targets
where in fact exactly the same and

when we look at differential expression
lies and I see less than ten percent of

the differential expressed
August where the say yes but

also if you look at the wanted a specific
only to P.D.F. X. you can see that

there seems to be a dramatic reactivation
of D.N.A. damage dependent B.

fifty three of those is control or
sort of program cell death control and

there's something that is interesting
because in this that was self

which undergoes somatic if you take
into the mutate their own D.N.A.

you want to completely block any
damage response because otherwise they

would kill themselves or
they're going through selection for

the immune system so
this basically is a drug that seems.

To be able to reactivate that mechanism
and that may be why it is in fact

the French reactor and there's a number of
other could change so this allows us now

that we have put all these things together
to think about a complete different way

in which we can go in terms of
translational research and development.

You see this pathway this is a path where
the normal is followed right today for

bringing drugs into clinical trials
first you have to find it compound and

have to go into medicinal chemistry and
refine it you need to understand

the mechanism of action you need to
understand by marks that can be used to

identify the population but also to follow
efficacy within the population when

you have all doubt which is six to ten
years you now can go in clinical trials so

a group of senators
including my center Columbia

received our money to see whether we could
do this in two years and is include two or

Strieber of the broad John Maine and
Michael White.

The Southwestern.

Scott Powers and Cold Spring Harbor and
this bill Hanna got a fiber and my my my

center of Columbia and the idea was can
we use this kind of function ation and

system biology approaches to do all of
these things are once in two years OK and

we actually signed up to do it for
three different cancers in two years which

we think is crazy but actually the results
are already coming up after only one year

and they look very very promising starting
with that you see resistance and also for

the G.M. And so there's a paper that
just came out this is not a peer

reviewed papers peer reviewed but
it's commentary basically.

The P.I. of the center co-author
of the century explain

how can you this type of approach is
to very very quickly identify all

the major ingredients I can drive
into therapeutic type trials.

And the way we're doing at Columbia is
by coupling pulled our nice screens

around to five potential master
regulators the system biology approach

evidence integration to sort
of to find a combined list and

then screening a small molecule libraries.

It's tens of thousands of compounds
twenty five to one side and

you see there a particular molecular
signature or a particular phenotype and

then combining these molecules with these
mechanisms through such as Marina and

idea and
just in the last slides on a show that we

looked ready with were Schreiber for
composite can inhibit such a beta so

we saw three integrated into five thirty
seven compounds that binds a tree very

specifically using and you can mistreat
came out of this tribe or a lab and

this shows kind of the synergy between the
what the virus sensors are doing and we're

now testing them using cells that have
been instrumented to report differential

Octavio districts in fact so they had
a conclusion and sort of reflection for

you to carry away from from the stalk and
I hope you know I haven't lost too many of

you because obviously there was a lot of
material here but the idea is the art and

I wanted to really present this stuff
because it it if you miss a piece you miss

the global story and one piece of
the global story is that the current

emphasis today is on Gene Harbor genetic
alterations this is great because

you can find these genes by sequencing or
by genetic sequencing

Unfortunately these genes may not be the
best entry point for therapy number one if

you have a deletion very hard to deal with
a gene that is deleted using therapeutics

number two if it's amplified to be example
of a very small subset of population so

it may work but it actually if you hit the
first second and third most frequent want

anything after death start to be
in the two percent range are you

going to develop a drug for something that
hits only two percent of the population

with a disease second one is that current
approaches to by market discovery should

be revaluated And the reason is that again
these issue of going down with the air

PLA airplane crash analogy of
looking forward not bigger is better

in some cases shown you know the genes
that seems to be the most sensitive to

determining the subtype are in fact
very dual differential Express.

The third point is that when we look.

It's much better instead of starting with
a Jew OS to end up with a Jew OS once you

have the functional data you can now walk
upstream pathways and look for lesions or

alterations or even variances that
are normally an error in the population in

a much more targeted directly way within
the pathways and finally this is something

that we're working on right now with this
we have project called a thousand drug

project which is a misnomer because
it's not needed one thousand or

more drugs but it basically says
that what we want to do is identify

a repertoire of model small molecules that
can hit really different and orthogonal

targets within the human genome right
now we're hitting with all the entire

pharmacopoeia of drugs that we have less
than two hundred genetic targets and we

tremendous side no off target effects if
we could have about a thousand compounds

that hit distinct I guess we would be
able to combine them very quickly and

essential drive through combination there
so let me just thank a number of people so

these are people in my lab and sort of
computational components of the lab

an experimental component is
our collaborators to your

account of far as lab of our own and
that's relevant all the work in G.B.M.

also we can all dopy clear
about a shot in my question for

the work in prostate cancer in
mice the store Schreiber's lab for

the again working G.B.M.
MICHAEL WHITE working line on cancer.

Funding sources Thank

you.

So.

This is.

The right ones.

All the.

Way.

Very low so very very non
differential expressed right.

So.

We're.

Right.

Here.

They where that what the point I was
trying to make was a different one

was that was sufficient for the idea for

the reductive is that was the four
thousand I can tell you there was nonsense

distinguishing if you can see where is
the tree were also not of so the word for

two hundred fifteen hundred there
were non sophistical significant OK.

For Scobbie rate of five percent so this.

No I mean is no not a test because it has
obvious have to correct for my firm or

discover a typical dislike things but
that's not the point really the point

is that you would never go to the four
hundred gene to validate it if you

have a result from the French expression
out OK and you would never go to see B.

and start three if you're looking for
more Geneses And in this

is due to different supply because he has
never been related to cancer before so

if you use a typical approach is even
if you went through the list and

say these are all the regulators you now
go from two thousand to maybe are three

hundred you still will be at the end of
this you still would probably overlook it

OK while if you use the approach sort of
protocol virtual programming approach

it actually comes right on top and
on the protein level

in the midst of chemistry can see
this differential expression but

the other of the gene expression level
there's no difference or expression.