It's good to be here.

And I mean they've been
a joy to take no for

I think twelve years and this is the first
time I even dark at the break for

a seminar so much really nervous it's good
to give it up when you're nervous and

because it's supposed to get the best out
of you know maybe the worst we'll see so

I want to tell you do you are serious of
projects that we've been carrying over

the years really I mean this this is work
that is stretching maybe I'm going to see

over the period of seven or eight years
now and all the work that I'm going to

tell you about this work that I've done in
collaboration with John McDonald on it so

it has to do with ovarian
cancer in these workers really

branched into other projects over time so
we do things that we've learned

from ovarian cancer we've we've done work
on prostate cancer and now we're starting

a few new things that we had never thought
we were going to be working on before.

This collaboration with John also ended
up branching into a collaboration with

College of Medicine with my team at Sauk
Hussan expert also you know Van cancer and

Marty's group has developed one of the
first in mouse models of brain cancer and

for reasons I'm going to they're going
to be very clear in a minute having

a controlled well behaved nice system for
looking at the C.S.

goes a long way OK we all like
to do these things with humans.

Expense with humans who only asked us this
samples that you have access to which can

be quite limiting in this case so before
they start telling you're out of a in

cancer I wanted to give you a quick update
on a quick preview I am lucky to have

Matt Doris and
they've smiley here Smalley here from.

Our new center George at their beck and
well it's not officially center yet

let's call it a color we call
a decision must be comedy center or

core it's a new set of facilities
that we are setting up in the new.

Even the one building and they're
going to have this scene proteomics

makes to the work on both target is.

Experiments and
that I'm going to tell you or

the study didn't until you do
experiments in our own research but

I wanted to tell you that this type of
thing that we've been doing as a research

project now they're going to be something
you could go to the facility and

say hey I mean pristine doing these
what what can we plan together OK so

the final space of the facility
is going to be in the basement of

this is not going to be ready until
I keep updating the slide OK so

where does it say now OK I just said
beginning of twenty sixteen because I want

to they did anymore about roughly five
February twenty sixth seen and it's

a really nice space where of a lucky very
lucky really beautiful space in the B.B.

but right now we're operating on temporary
space that was given to us temporarily for

six months it's twenty thirty
forty Navy and the manager of

the courses here David So he's the point
of contact but I'm happy to help you with

the See if you have any questions so we're
very lucky that they see the invested

in some instrumentation for
doing this work we have

in a street of instruments these are all
must try matters that are used for

measuring different molecules
in different types of samples.

That have different kinds of MUST backs
because we want to return in either

proteins.

And so there they are these like
a bike OK to her road bike or

mountain bike they're both bikes but
you're not going to go we're a road bike

on a mountain trail you can try but
it's not going to work OK.

Face one of the deployment of his
ferments is pretty much owned by when

they see Mr Here they're pumping down
they're being tested as we speak and

then we have a second face of instruments
that we're thinking about what.

Capabilities who really want to add to the
core asked we have customers coming in and

basically saying this is what I
want to do this is one of OK so.

Do a lot of work OK So we've done some
work we as I said but of these development

we entered into collaboration with thermal
because we wanted to save some cash so

we see that about a million dollars by
entering into this collaboration and

now we are partnering also with Emory
because Emory has a big cystic fibrosis

center and that center is very interesting
you seeing the course here and

some of the couple really here on things
that they don't do it Emory OK so

just try to complement what we do here
was what's been done at Emory OK So

with that I wanted to give you a preview
because they know they're moving if

you maybe interesting to some
of these experiments before but

I'm here to tell you about double
our mix and this one of the new were

all mixed feels and
you know that all these all mix feel so

really enabled by new technologies
OK ginormous was enabled by the way

I really did to sequence genes OK And so
that it's really interesting to really

look deeply into how these fields emerge
because at the very beginning this sort

of we see a lot of research in analytical
chemistry developing the tools and then

the tools become very mature and they're
taking away a biochemist on biology and

the analytical chemist jump onto
a new thing OK so then you came

proteomics was was really enabled by
high throughput must be crummy tree and

soft only sation must pick dramatists
in the ninety's and that's pretty much

it's a very sophisticated very complex
set of tools but it's doable and

it's relatively routine depending on the
expend that you want to do some expense or

routine some are much

more complex make that a lot
mix came after that and

I am going to say I'm going to venture to
say that it's moving faster a number of

the are mixed because the terms from a
double omics are similar to the proteomics

with the exception of
the by informatics part.

They are buying from my text while
it in mid-air alarm it is still

where proteomics was fifteen years ago so
it's still up in the air lots of

things changing different groups coming
up with different tools and so one but

me that alone makes as the name
imply ease the system level or

network level measurement of metabolites
some people call Global middle nomics I

don't like to use the word global because
as you're going to see we never measure

all metabolites known to man OK we never
measure that we cannot measure that all at

once yet OK so it would be too ambitious
to say that we are doing that but

we're trying to get there OK typically or
operationally speaking we're looking our

relatively small molecules in the range
of fifty to fifteen hundred Dalton OK So

this involves sugar slippage some you know
acids you know steroids you name it and

these paper by Skolnick group that
was published years ago I'd like

this model off of MIT that were laid out
to ration because Jeffs group was really

talking about him it that will
make al to ration a sick way that

a chemical engineer with think about
metabolites going up and down so

if you have a chemical reactors with
reactions that produce a certain

chemical or deplete that chemical
meaning in science that may be up or

down regulated in deceased or something of
perturbation then metabolites may go up or

down you can accumulate or it can deplete
that modality and I always thought the use

of a simple very very elegant model off
networks and how many go up or down.

The metabolism her Scilab of interesting
information that may not be obvious

in the protium or the genome or
the transcript Tom so much more an instant

snapshot of what's happening in a given
system as such you've got to be very

careful because that instance snapshot
can be altered if you mistreat

your samples OK if you don't process and
careful if you take too long to process

them it's going to change because
it's really a much more instant rip.

Isn't ation so experimentally speaking
it could be a bit trickier than perhaps

experiments with proteins or
even genes OK So metabolites you going to

be very skeptical of the results
until until shown otherwise and

I told you I was going to talk about
ovarian cancer I don't think you really

need to explain my why ovarian cancer is
important it's really a deadly disease

it's one of the leading causes
of this thing women of cancer.

Perhaps the most important thing is that
each of us at every stage she has no

symptoms and it's very difficult if
not impossible to catch it early and

my my grandma actually they are for and
cancer and she didn't know you OK So

he has no symptoms so they only it's only
found in the early stages by chance and

he's only find really by chance and them
if you found it early because they happen

to do surgery and they did take early
stage cancer survival rate is very high

OK so if you find your early
the chances are of a high so

we want to try to find it orally and
we've been working with John McDonald for

years now trying to find ways of
the acting early stage ovarian cancer and

ideally what we would like to
the discovery say blood test we take blood

we should that into some sort of
instrument doesn't have to be an aspect

I'll be happy with anything and
it says green or

red OK But you know it's never black or
right there so

we said range of your situation so
we're going to talk about that

like with many cancers ovarian cancer
is not a single type of the Cease

it's actually the most common type of
ovarian cancer it's called the serious in

serious serious so when it's serious
popular it tends to Mr metastasize and

it's pretty much deadly if it's not caught
early but there are other types and

some of these other types are much more
rare they're not really understood

if they're the same type of cancer
of the serious popular but typically

the development hypothesis is that
ovarian cancer happens because of the.

Repetitive grounding of their own very
surface okayed in the menstrual cycle and

during that process the mutations that
trigger cancer but that hypothesis has

really been challenged by many researchers
including the idea that maybe a brain

cancer actually starts in the fallopian
tubes not in your body surface OK so

that's why that's important because thirty
slides from now I'm going to show you some

results that shows something like that OK
I'll remind you the current standard for

the taking over in cancer is the Michael
called CA one twenty five that it's as

good as pretty much flipping
a coin if you put it mildly So

in some cases it's useful if
patients that have mutations that

you know make them more prone to ovarian
cancer in those cases it's used but for

the general population in general
it's not that useful OK So

many many many doctors don't
even want to do it because it

shows that it's elevated it doesn't
really mean anything where you've got to

really track it over time OK so if you
track it over time and sign Lisi going up

that may be indication of some
problem Nonis earlier in cancer but

some soft ovarian problem so I'm going to
tell you what markers for of an cancer and

am pretty much the community of
biomarker discovery has agreed that

detecting one biomarker is just not going
to happen anymore all the diseases for

which one biomarker is enough they're
all done OK we're going to need panel so

via markers we're looking at
a very complex disease and

we're setting the bar very high we want to
do early detection so we really want to

look at molecules in a group of molecules
maybe four five six then going up or

down OK so

that's what I'm going to be telling you
about the recent mathematically it's very

easy to understand what the right that has
more discriminating these community power.

If you only have one biomarker and you
have two populations and these biomarkers

not they specifically are sensitive
you see in this case there's a lot.

Overlap between the populations right and
he's going to only in the most

extreme cases if you have a second mark or
somebody may propose

again along this axis there's quite
a bit of overlap between worst but then

you can combine these two way markers you
can just say it in this case is a trivial

mathematical function it's a straight line
that combines the abundance of A and B.

if you combine those you can basically
say OK this is the dividing line and we

deceive a line I can actually separate the
population as well if you can imagine D.C.

in eleven they mentions that's
basically all we're doing OK so

you have in multivariate plane that
can separate populations OK that

say the it's a function OK sometimes we
call a disk or we just call it a score for

simplicity but OK you don't have any
breakfast so maybe I should have one year

old that's coming out OK The next
slide is a little GROSS OK so

we would like to do really stage
the direction of an cancer.

And we would like to do that in humans but
because it's so

hard to find early stage
someplace in humans.

In parallel with the work that we've done
humans we're also working with a mouse

model of ovarian cancer and the reason why
it's important to look at a mouse model

is because you can follow the C's
progression in a very controlled way so

together with Bayer and with John McCain
And we've been working on these mouse

manual that it's basically a double
knock out mouse model or worse the.

Beat and believe it.

Beat anything suppressor gene and
dice or survey glittery gene and

when you suppress these genes in the
Philippian to use Actually what you see

this is the whole reproductive system of
a mouse OK this is the uterus here these

are the Philippian to you these two
little glands are the ovaries and

in this case the system is clearly benign
OK this is a cyst and this is an early

stage two more that has not yet
missed that size so basically you know in.

A typically around six to eight months
these mice will start developing ovarian

cancer you don't know exactly when but
if you sacrifice them you can find out

if they're early stage or a late stage so
when we do this we started studying these

mice trying to look at marker seeing
blood of these mice and after ten or

eleven months the tumors actually
spread metastasize mice die

every single time OK So this is
a pretty pretty good model of ovarian

cancer it mimics many of the molecular
features of human ovarian cancer for

example beast mice have elevated
C A one twenty five and

many other features that you are seeing
human So as far as we know it's a adequate

representation of what you would see in
human patients so what we said to you is

we said to the limit of our lawmaking
investigation we said to say OK is there

any alteration in blood from these
mice at the very early stage of

the Seas is there anything that we can
take that is going up or down that we will

be a good indicator of the Seas typically
what we're doing right now is we're

applying these very complex and
workflow we had to give it a name

everybody has to have a name for
these things we call it music so

it's a myth our long live coverage
water flow it's very complex so

basically they get a blood sample
we precipitate the proteins we end

up with serum sort of serum sample
receptor precipitate the protein we end up

with a metallic extract you can do that
precipitation with different servants and

you end up with different extracts
of the from polite These So

you're really doing a fractionation
rate to begin with and

you can have a more polar fraction and
less polar fraction the less polar

fraction we apply a mix product all
the expertise of Merrill's group and

Cameron solid Scruby to basically
specifically for Libby it's.

These a fraction the more
polar group we split into and

what we're doing now is worse.

If you do any Mar This is
together with Edison at U.G.A.

and we're also splitting it to do is
separate type of mass spec just for

the polar fraction so
you see this lot of work involved in this.

The trick like I told you before comes
after you combine these huge amount of

data how do you do with it so
Julia Quranic and I had a project for

three years and I always joke that we
collected data for eight months and

we spent the last two years deciding what
to do with that they did because it was so

much so complex so careful what you
wish for in that case we actually had

me double law makes must begin in
a moment the longing to blast but

the omics data on the same system and
he gets so overwhelming OK so careful how

far you want to push these things but you
need to find a way of combining these and

obtaining what we call features features
are either pairs of mass to charge and

retention time or you know the frequencies
in the N.M.R. and then identify

those mentalities which are going up and
down so this whole protocol is where we're

using right now on systems that we already
done some pilot work in the case of

the mines we did a study like this
because we didn't know if there was going

to be anything in early stage.

Mice OK so I'm going to show
the simplified protocol which is to

take serious samples we
precipitate the proteins OK And

this is actually sound simple but you
go you got to this is where your O.C.D.

has to kick in because garbage in garbage
out trust me if you don't do this right

everything else is a waste of time and
we've been there OK we've spent a year and

a half on some pills that
were a waste of time and

then at the end you talk to her
collaborators and you're like Are you sure

that the cancer samples spent
the same time in the center future.

And you see them pass for a minute and

they're like I didn't realize that that
was so important that's it it's a year and

a half of work thrown away
because if you don't.

Keep things constant can be any
variable that will distinguish any soft

confounding factor that will distinguish
between the two populations and

you don't want that OK you want to
me to minimize that once you prepare

your sample you go into a must pick
trauma or a couple took a month or

a few Typically you'd be able to see two
separate and those metabolites that's

a pretty important process like I told you
before key to this process is the ability

to take metabolites in the liquid face and
put them in the gas face with a charge

must be trying to work with
that works with ions and

we use of a quill technique called Electra
spray or any station that basically

this was not invented for me double
omics the it was actually invented for

painting cars OK So electricity was
invented for spraying paint on cars but

somebody very clever said how well we
use it to make irons and he works and

then you can make a flowing you look with
coming from your chromatographic column

and applying a potential charging
the droplets of a high voltage OK and

that makes ions of those

science go into the separator if in
this case we use a time of flight and

time off like this one of
the most elegant techniques for

must pick from a tree the way
that it works is that.

If you start with a mix let's say that
we're all ions OK we're all holding

a charge we're alliance and we need to be
separated we all have different masses

my mass is conflict growing it
will have different masses and

we're kicked into this chamber OK we're
kicked and we're all given the same energy

but because we have the from mass to
compensate for the same energy we're going

to have the from velocities OK High Mass
is going to mean low velocity and

vice versa OK So the way they were
actually is that the instrument

basically injects the ions OK
inject the audience here and

the ions start separating So
this will be these would be heavy I.O.C.

to the light Iowans in particular
the stable things from and use this what's

called the mirror reflection where the
ions bounce back and hit that effect or

the reason why you need these mirror
is to keep the instrument comp.

He would make it pretty long OK So
this is a time of flight of fun and

this process that took me three
seconds to scroll through the slides

have been seen microseconds OK so these
have a fast instruments keeping up with

the chromatography that is happening OK so
I'm going to go quickly through so

you separate the you separating the U. B.
and C. and you further separating your

must pick from it or the computer
while police keeping up with this and

detecting everything they should be
in this getting the detector and

you're going to end up with these
dimensional data which he's

chromatography in one axis a mass
spec in the other axis OK for

each sample you're going to extract from
each one of those from that they are going

to extract the peaks and you're going to
align the peaks so you know when you want

to make sure that you're comparing apples
with apples that's step off extraction and

I'm in an integration it's also
where all the problems happen so

it's full with pitfalls they
important that it's done right but

once you end up with this
massive spreadsheet of.

Features we call them we don't even call
them mentality because we don't know what

they are you need to start
figuring out what those are.

If you go to a conference where
somebody starting on MIT double omics

some people will claim I am
detecting a million metabolites that

wrong OK there it
effecting a million peaks

those peaks may not be metabolites
they mean contaminants from the V.W.

are methanol contaminants from your column
OK so you go you're a colorful and curious

curate you're eight that data to make
sure you don't speaks are not known for

those peaks are not in the blank and
so on OK So he gave a code for

with that step but once you go over
that step and you're sure that he's done

ask Lena's you can then you're ready
to go and talk to your bio informatics

colleagues and start playing and trying
to see East they're a function or a.

Is there a set of metabolites that I can
use to separate let's say early stage from

control my eyes don't meet the answer
maybe not sometimes there's no way of

separating it sometimes it doesn't
work work OK so we curate data and

then we do multivariate analysis and
then what once you know you don't meet

the lies you're going to identify them and
I'm going to show you how to do that and

then you need to map them back into
pathways OK so you go from blood

all the way to pathway mapping So let me
just show you some results from mice so

the first thing that we did is we ran
a set of minds through all this workflow

actually scribed and we had early
stage mice and we had a late stage

control myself case so early stage
mice are the red symbols Let's take

a shot of the green ones and the blue
squares are the controls if you take all

the data that you get from there must be
a trauma and you do something relatively

simple such as principal component
analysis and this is the B C

one versus B C two this is
what you get you see the and

the late stage minus separate relatively
well from both the early stage and

control but the only station control
don't separate Is this surprising

No I tell you the early stage
has no visible symptoms So

overall it's very hard to detect OK So
overall it's very hard to detect So

this was a little disappointing but

we're like look wait when I
mean it first of all P.C.A.

is trying to capture these things are the
most to different Ok so late stage is so

different these mice are extremely sick

that it's obvious that that's different
from the other two sets OK so

that in itself confound say a little bit
the confounds a little bit the analysis so

when we do this we took away the late
stage samples from these dataset and

we stuck to the early and controls and
you still have you seen P.C.A.

we use what's called a supervised method
so we basically split the data into.

We used a training said test set
the training said builds a model

build a function that separates and the
test that we pretend it's an unknown and

we see if we can classify OK So that's the
way it works and we used all the data for

this OK we're doing the right it is set
off nine hundred thirty four features like

I told you I mean there are more than one
hundred thirty four known metabolites So

by no means we're looking at the whole
metabolism OK it's a good fraction maybe

I'm going to say a tenth you
know the bending or you know but

this is all negative IMO data so we're
specifically looking at molecules that can

be proton eight so let me show you what
happens when you say supervise fashion so

we first train a model and of course when
you train the model it works very well you

can separate early stage from controls
perfectly when you train the model OK And

this is using all nine hundred thirty four
features but then when you test the small

it doesn't work so
well you see how there are some blue

squares over the decision limit and some
red circles below this issue are limited

so that means that we're
misclassifying mice and

you don't want me solidified
samples because it means that

you're telling somebody you have
ovarian cancer when they don't or

you're telling somebody you're fine and
go home and they have a then cancer so

none of those situations is really
something you want to be in OK so

it's pretty challenging to do this so
what can you do well.

Computer scientists are to have
figured this out before OK so

in the same way that when you grow
when you open Netflix Netflix kind

of knows what you want to watch next and
that's I'm not joking actually the person

that invented that if the person
that this this work with us OK so

it's the faculty member in computer
science Alex Gray he's he's he's

millionaire now because he figured
out the network thing you know but

this is called Machine Learning OK so
you can teach these models how to live and

what are the best metabolites for
separating these things OK so why.

One way of looking at it is by using
what's called a genetic algorithm and

I'm sure many of you know our genetic
algorithms if you're done some genomic

work it's used all the time the way these
the following we're going to take subsets

of these one hundred thirty four let's
say we're going to take sets of ninety

we're going to make random sets
of ninety let's say two hundred.

And we're going to try them dusty
said work to separate No Yes maybe so

you're going to rank them and you're
going to see which sets were the best and

which sets don't work at all or
no mediocre you rank them you throw

away the bottom half and
you keep the best you shuffle them and

you shuffle them using techniques similar
to generating mutations so you have

mutations you have crossover and things
like that and you start over again so

you start sets off night ninety and
you end up with smell of sets but

they work much much better so this process
is actually repeated over and over and

over again you can take this to run OK so
we did that with our amount of data and

we came out with a new set so this here
is thing you said so we went from nine

hundred thirty four features to
an optimized said of eighteen features and

in this case we can probably safely start
calling them at the lights that actually

separated Well not only
in the training stage but

also separated in this case exactly
separated in the test face and

the reason right works so well even in
the test faces because these are mines so

it's a relatively homogeneous population
soft samples OK I'm going to show you

broke with humans that's all right so
well OK but

in mice you sometimes you can expect
these and this is all nice and

great but now we need to know what
those eighteen are because it doesn't

nobody wants to predict on Peaks people
want to know what are these molecules and

do they make sense because if
they don't make sense then.

We need to start over again may.

Sense means they do mapping to
map a biological pathways that

are known to be altered in cancer and
particularly maybe nowhere in cancer OK so

how do identify these so this is if you
have taken an organic chemistry class you

go back to your guinea company
going to chemistry class you look

at your chromatograph this is one for
a control sample this is one for

an early stage sample and we've
a tempting to start to look at peaks and

small differences but careful because
you're not just interested in differences

you're into seeing differences that
are covered with early stage cancer there

could be many differences OK between these
mice but you just want to look at the main

difference is a similar fact if you
look at this blood here you see that for

example early stage mice in this direction
have a lot of variance but this is

not correlated to early stage disease OK
so careful because this is not the range

that we're interested in going to see in
this orthogonal various OK so you look at

the chromatograph and then you look at
this peaks in the chair last changed

I'm just using one here as an example we
should we saw that this mass to charge

ratio for
fifty two point two seven seven nine

were US altered you see here two
traces one for control in blue and

one for early stage in red and
then you look at them aspect from so

you're really drilling into they do you
look at them aspect from under these and

you see OK this is that big that is
changing and with decent mass here these

accurate mass many times you can obtain
a little mental formula for that molecule.

And that recently is why we rely on really
high resolution instrumentation because we

want this number to be very reliable and

we don't want any peaks to
overlap if you have all wraps.

You may be looking at two molecule set
ones and that's one of these stored all

that picture OK so that's why we need
high resolution measurements Furthermore

the mass alone is not enough for
you if you select the smaller.

The time of flight has a way of
selecting a specific ion and

you fragmented you break it into parts and
you measure the mass of the fragments and

then you start putting the special
together if you're lucky

you're going to be able to search the
scene is that a vase but that always has

formidable omics are not there yet they're
growing but they're not complete so

you sometimes you may search without ever
saying get a match that's great but if you

don't get it much you're going to have to
look at what potential molecules match

these filament old formula and then see
if that molecule with match tandem aspect

OK so it's a pretty this process process
so it's some tedious that you're going

to understand why show the stable so you
know they teach you in kindergarten and

of power point not to show
tables the stable to gas

a year of work so I am going to show it to
you like you don't know because it was so

much work OK We're very
proud of this table so.

By no means I want you to remember
everything in this table but

what's really important
is here in this column.

Identities of these molecules metabolites
they have different degrees of

confidence some are probably
wrong some are probably right

because we cannot do all of them some
of them are very low abundance and

on those you really cannot do
reliable fragmentation OK So and

some of them we have fragmentation and
we have masses but

we cannot do chromatography well
because there are no standards OK so

they once are a barrel that for
example here one molecule that was

really interesting was Billy Rubin OK Bill
a ruling was changed in this in this

panel there are many bits forced by
the really nasty dolls is somebody lasted

some trade release arise but
there are some molecules for example any

OK Any Any So
this is by no means completely than OK but

if you we did then that the way
these if you go back and you my.

Pathways you see really many changes
you see changes in Leap in that Daoism

by the way Blue means increased in
the early stage red means decreased

in the early stage you see changes in
the here in catabolism there's Billy Rubin

here downstream it's well known we
had detected this before in Leith

altered by last admit that we listen so
lots of things going up and down OK So

this is really interesting but doesn't
really answer all the questions it's

better for because it means that there
is some sort of signature of early stage

of an cancer in mice but you know we're
not going to take this to the clinic yet

right there's a lot of work to be done and
one of the things that we particularly

interested in is how do you really
improve this in an identification process

how do you really improve it and
that's important because any biological

conclusion that comes after this
is based on the kofa lady and

you'll be surprised how many papers
are out there where they are treated they

likely OK you do you shouldn't do that and
one of the things that

we're trying to do is to measure some
other property of the molecule not we

measure the mass we measure
the fragmentation those are intrinsic

properties of the molecule and
we measure the retention time

in a column that's not really a very
stable mission because it changes between

different even between different
batches of columns if you then H.P.L.C.

you know what I'm talking about you
know you can have you know you put

a new column and retention times change so
it's not very reliable so

we're working on what other things
can we measure let's say for

this or any more like you
look a this is generally for

any molecules to improve identification
and we're using a technique that is not so

well known well you probably seen it but
you don't realize that it's called you

mobility spectrometry So
if you've gone through Hartill Jackson and

you've actually been searched with one
of these or even seen one of these so

you know when you see when they do this
rubbing on your fingertips and they're

looking at you that's what they're looking
for explosives or narcotics so I would.

I go through high school Jackson I'm ready
to say look I'm a chemist it's OK for

me to have chemicals in my hand
which is probably not OK but

I am really dismissed are basically they
Smiths back there OK in that corner that's

what they what they use and but they can
also be combined with mass specs when I am

ability is like a guy's face
electrophoresis OK so you make us and

then the science fly down a tube and the
see gas that floats against the ions and

separates them what's intrinsic about
this process molecules are separated

based on a cross-section and
that's the shape of a molecule in

the gas face OK and that's very intrinsic
often doesn't change between labs so

if you can measure that and you can
match the cross-section you can have

an additional way of saying yes
this is the molecule I think it is

the fellow may weigh instruments This is
called a drift you are you more reader

spectrometer these were developed
back in the sixty's in physics OK and

then they made it all around the world and
they're even commercial OK but

the theory for these this is what's
called a mason shampoo equation

the Coalitional cross-section often
a phenomenon is basically proportional to

the drift time in this you proportional
to the electric field in that you so

this is like an electrophoresis and it's
inversely proportional to the length right

so the ear over early is
really what matters so

we're trying to find ways of using
omega as an additional parameter for

them to find metabolites and this is
something that we're building right now so

we have one of these instruments
only trap dramaturgy this of a cool

type of mass pickle tell you
more about this in a second but

we're combining the mass back wheel and
mobility so

after the crime of the graphic separation
you get I am really the separation and

after that and the cross section and
after that you get the mass to charge and

after that you defragment So now you have
redemption time coalition a cross-section

mass and mass of the fragments for
eighty is my.

OK.

Getting back to the early trucks all the
traps are really changing the landscape in

terms of Master cometary And

if you paid attention the new facilities
in even we are going to have or

we trap must be travellers all the traps
are for you to transform aspects so

basically you take your irons you should
them into the store beat up on this or

get up to you know these are this big Ok
so it's not the monster instrument that

must analyze or so it goes into the orbit
trap and they are still eight and

these oscillation frequency is
proportional to the mass to charge so

if you measure the oscillation

over time you apply the freer transform to
that you get the frequency of oscillation

that that's the mass spectrum
basically of a they wrecked OK So

these are very high resolution instruments
very very high resolution instruments and

can separate very closely related
molecules that's why we want them OK so

in this case where you have only one kind
of molecules you get a sine wave but

if you have to make sure you get
a mixture of sine waves that you the.

Fast would transform this whole thing
happens behind the scenes OK so

you don't have to do this manually.

OK So I show you this list of
metabolites and the question is.

Are these coming from tissue or
this is related to the tumor

what is the connection we what's
happening in the ovaries so.

The other thing that you can do in these
experiments is you can look at the tissue

itself but you don't need to do that
no hormones are nice and extract

the tissue you can actually grow and do
mass spec they rectally on the tissue so

there are many worse of the we must pick
from a tree imaging our growth facility

here in the east very good at doing manly
imaging Malise a technique that works with

a laser and a matrix so the laser
shoots on their tissue sample and for

every spot you get him aspect room so
if you raster the lesser then you get for

every spot in aspect from and then you
can say OK build me an image for M.

over C four fifty two and he said
virtually I'm just not really an image.

But he will plot that mass over X.
and Y. so

you can do that routinely these
days OK Manley's of a powerful but

that's not the only way for example
there's other techniques like seems

secondary I am a spectrum a tree which is
actually a judge they cost the same C.

instrument is used for materials for
imaging materials air but

there are other techniques that are faster
and offer complimentary information we use

a technique called SE Bessie's
very interesting I call it and

slightly works like a pressure washer
if you are washing pressure wash your

driveway you are shooting it look with
Jed at the tissue sample so you're

shooting a liquid at the tissue sample and
the droplets are extracting metabolites

from the surface and bouncing into the
master trumpeter OK So it really works for

a mechanism that's very different than
Molly and that's why they provide

kind of complimentary coverage in terms of
metabolites and things that you can see so

let me show you results OK So we typically
apply several techniques we take it to

shoe sample we use the cord here
to he started to court to take

a slice of the tissue sample with a crayon
micro Tom we put it on microscope slide

we throw it in this cicadas sometimes we
wash it depending on we're looking for and

then if we're going to do manly we use
the I.V. we call facilities here and

they've developed a system
to deposit the multi matrix.

And we use them only instrument in
their facility to image by Molly but

we also do Dessie in our labs and this
it doesn't require the matrix it works

a little bit differently OK So
we are going to have two sets of images.

This is word the whole reproductive
system of one of the my

mutant minus an emerald Eastleigh looks
like OK so it's right because it's

been a really coated with the molding
matrix the millimeter is the compound that

absorbs the release of radiation
OK I don't know if you can see.

It's this is the uterus
the one fallopian tube here.

The other one here one already
here another already here a big

late stage two more here
filled with chili and

there's a little early stage two more
there so we actually sacrificed many

animals until we found this particular
sample because we wanted something

where we could see a comparison of early
stage and late stage in the same tissue so

we got very lucky to find this one and
we imaged these both in Pasadena native

I am on but I'm just going to show you
one set of data we have too much data.

This software part of this is not so
simple simple it's quite tricky so

we've been working with may one for
quite some time now on developing software

to take the image to take the data from
the mass spec and assemble an image so

we came up with the software tool
called spect this is actually online so

you can upload your master data and
it will make the images for

you you say I want to look at a specific
guy and I would make the image for

you so you don't have to have
a powerful computer to this yourself

it's web based it's open source or
you can modify it if you so wish to do but

the main thing that I like you study
you can do quite a bit of multivariate

analysis it has a routine called an M F
non-linear Matrix it's like it P.C.A. for

images basically and he can show you what
are the main changing in the image you

generate a lot of the images you generated
at running for every embassy so if you're

going to look at these images manually
it's going to take you this OK you're

going have to go image frame should be
like this is interesting let's say your

side and so on so this will do that for
you OK and he works for

any soft technique.

So some images this is the reproductive
system and if you can tell that he follows

the shape and what I'm plotting here
is the end then method image OK So

this is the changing the most
are not a single line OK So

you see that there is quite a bit
of changing in these two or more.

That don't overlap with some
things changing in the uterus and

in the early stage tumour OK so it's
not a simple picture it's not black or

right things seem to be changing in many
parts which goes back to the theory that

maybe things not don't necessarily
originate just in the ovary

maybe there are some that are maybe
migrating to the ovaries and

that's what's contributing to
the relevant offer in concert and

these type of images are called component
images because they really combine several

lines at the same time but you can also do
just a region of interest image where you

just select these part of the data and say
OK show me all the ions that are here so

if you want to look at diet all these
all the science in these particular

particular tumor region are shown
in this mass spectrum and

you go through the same in difficult
process I was telling you before these By

the way early bits which is not surprising
OK and I show you a list of another

list there were very proud of because
it took a long time what those are and

I don't know that we used to techniques
and they're really complimentary OK So

this is the man the image that we have
seen with Cameron solid seen the I.V.

facility and
this is the best See image and

you see obvious some obvious differences
look at the tumor region here and

they do more region in the Dessie image so
they're really different OK So there's

some other differences for example if you
look at the lower mass range the best

the data is much better at the low must
range where manly tend to have a lot of

noise because of the matrix
in that range so

you see that ions here that
you don't see with Molly so

we use both kinds of data OK which
makes things even more difficult but

more complimentary on the other hand look
at the Molly data for the this early stage

the ovary so the Ori and the early stage
tumor lighting up like crazy MOLLY So

there's things that we see Molly that
we don't see in this vice verse.

All right long least but what's.

Important here is that if these are the
main species that we see changing in

early stage tissue there are limits
that once I'm highlighting here actually

are also changed blood OK so we went back
to their blood medulla mix they tested and

we look for those so there's
a connection between what's happening in

did in the two month issue and
what's happening in the in blood.

A Not for example this one was
one that was big for the panel so

Billy Roubini saw also something that
shows up very clearly indication of being

altered OK But you know there are not
a single way of showing the ration and

that's what's interesting there are many
ways of showing the celebrations there

are many semis there are six cholesterol
sulfate there which we've seen some

similar molecules seen the other panel so
it really matches them it makes data but

you know just to be true these you know
look at the number of peaks that we're

seeing in these images we're never really
taking five thousand signals OK so

this imaging experiments again they're
not global when it when anybody

uses the word global you know
that that's not there OK So

we're only really looking at hundreds
of metabolites in these experiments so

you do need to use complement a technique
so we are really gearheads and

we're really interested in
developing new technique and

we will bring a new technique for
imaging we use is a third way of

making ions which is it blasphemy and
electric blast my spark and

the laser OK So this is not ready for
prime time in terms of using pressure

samples but hopefully you know in the next
twelve years Callie invites me to give

another breakfast talk this would be
really OK but we're working on this.

You know this is so nice and great from
the economic perspective but at the end of

the day when we want to do is when I
was agnostic scene humans right and so

far what we've learned is that yes there
is hope in them it double long for

looking changes the early stage of
cancer at least in my eyes OK but who.

Right human so the work that we've been
doing for a long time with John McDonald

is really on directly in human samples so.

The challenge is to get
over ten enough samples so

the study has enough power to be
believable OK because nobody finds

early stage around cancer so we have
been collecting early stage samples for

a while and here you see
the largest dataset so far and

the paper got accepted this morning
actually in the you know we have heard

about forty six and
early stage samples and we each match them

two and controls they have to be H much so
you don't distinguish by age and

these are only details in case you're
interesting but they are Stage one and

two meaning that he has not
metastasized So this early stage OK so

we did this is on the H.T. solution we
did I want to save you all the details

are with the they similar experiment what
I showed you for mice with these all

limit our mix platform and I'm also going
to show you the negative I am more data

although we also these other types of
data and it works OK works very well so

if you look at all the metabolites all
the features it doesn't work very well so

this plot here which is the most important
shows you the number of features that

we're using and this is the percent
accuracy if you want to put it with so

if you use all modalities effected.

The specificity and sensitivity you
know sixty percent which is close to

flipping a coin that's not going to work
but then if you use a feature selection

mechanism like a genetic algorithm in this
case we use it more sophisticated approach

we use something called recursive feature
elimination support of vector machines

these are actually much more sophisticated
because in human data the variability is

much more complex to model and you let
it select there's an optimal number of

features around in fifteen right there
were you can discriminate they well so

if you take these subsets here you see
you Candy discriminate between control.

Patients an early stage cancer in
human samples OK So this is what

we want to do are we ready to take
this to the clinic I'm a pessimist and

I'm going to say not yet so don't try to.

Say that we have a diagnostic test because
it's only fifty patients I would like to

see another study to validate this with
let's say two hundred patients but

it's hard to find those samples so

we're trying to do that the myth that
relates involved again they go up and

down OK there's again in both directions
and we spend a ridiculous amount

of time also validating the small to make
sure that we're not fooling ourselves so

we try different types of classification
we try different types of what's called

cross validation cross ideation it's
interesting you example in an hour.

To pretend that you don't know what
they are you try to get them and

you say if you did right or not on the
back and forth and you do that over and

over again and the accuracy ranges from
perfect in the most optimistic case to

around you know ninety percent in
a more realistic case OK we want to get

better than that because ovarian cancer
is fairly low prevalence the seas so

you don't have the luxury of
failing when one case shows up so

you really want to have a specific in
a sensitivity to about ninety five percent

OK that's really what we want to get
in terms of what the metabolites

are it again this is in many an ace OK
there are many things that we could

in about five we spend months and months
or months trying to figure out where they

are we just don't know what they are yet
but many of these are limits so

there's Semite here and
so on but you have.

You have many interesting molecules you
have for example some small peptides.

You know if this is going to hold the for
a larger population we don't know yet

OK we're going to have to look at a larger
population and see if the same markers

hold because some of these markers
can help you discriminate one or

two some poles but not for
a larger population.

OK So this is where we are right now but

you know it's very promising
a way excited about this so

with that I'm going to finish this is the
group right now they were they showed you

imaging work it's mostly Marty Payne
the Ovarian Cancer human work

is by Dave Grohl who's going to be
working with Dave smally in the course so

he's going to be working on them
at that will mix part of this and

then did the rest of the group which is
getting a little bit too big at this point

and thank you very much again for
the invitation.

Thank.

You.

Yeah so
you know my limited experience with M.R.

is that it really highlights a different
part of the metabolism so for

example the world that we publish
we publish we truly Julia in

D.N.A.'s I think was last year if
you don't sound that were seen

in my we were not seen him aspect we're
not detecting them OK And that's because

there's a lot of overlap and suppression
system issues that you inject in there so

complex that some of the poor
ionizing metabolites just don't

show up but any mark because their mom and
so it was really surprising but

then if you look at pathway wise
they tend to map into the same

places which is really interesting but
the challenge is really to.

Join data.

So it's really different here where
people are really working on how to

join in M.R.I. big data and
not just from the pathway perspective but

also from even from the simple
classification perspective there are there

are papers out there I think
that's what people are I was

reviewing a proposal last week
everybody wants to do that.

It's not you know yeah
that's a very good point and

I'm I don't know the answer to that but
I need to look at that so

it's like an insult product Yeah it could
be because they're very very fragile

Yeah they're very fragile and
you know it could very well be that

that's the you know the way that we see
like the any safer conditions is typically

a compromise because you know you're not
looking at a specific metabolite and

you can optimize for a specific one.

They're not they're not.

You know they're not and.

Right right right no they're not and them.

You know it's something you ask your
collaborators what they've done with

your samples and
you tend to believe them but.

Yeah.

You know what that's a very good point.

Thank

you.

Yeah that's a very good question so
we tried to do that but you know.

The answer is this if you don't
have enough samples to do that

it's very tempting to do that and
you can do that for

other stations too but you know you go
from thirty six to or twenty samples and

then you speed that into a training on the
test said I don't want to go there I mean

used you start making hypotheses of
things that may not be true so you can

you can separate the stages let me just
say that yes you can separate them but

the station is also not necessarily.

It's science it's done by
looking at the tissues

something that is called stage to some
other person group called Stage one so

you know you need you need to
assign that class you know.

They carefully weigh which pathology
is don't necessarily do you know or

care to distinguish so well but
how would we would love to do

that.

So this stage most my understanding
OK of this is that it's Stage one and

two are basically a class of I when
they're not missed that metastasized

at all OK so but after that stage stage
three so really made a static and

it has spread into the abdomen OK so
if I show you some of the bigger soft.

In the minus I mean these mice have
developed with color society so

recent be exact soft blood and.

They die reversely So what kills
them you know the cancer has spread

everywhere.

Well yeah OK so
I understand your question so

when we call it let's take us all with
methodic OK it's not two more sites

it's made a study can we spread
the word to mostly other bugs.

In.

Blood Yeah there will be blood and

the way that we envision it
happening it will be the.

Leg A blood sample in the clinic and
this will be sent to a clinical lab OK and

the way that their work will be done is
the same way that neonatal screening is

done for kids where metabolites
are measured routinely So

you know a panel of depending on
the state about it in your thirty or

forty metabolites the techniques are very
streamlined you could do the same

thing once you have a panel
that is reliable and

you know the identity of those
molecules you could just say.

You'll have to change.

You know I mean it will be the will there
would be the women that went to surgery so

blood was taken before surgery and the had
surgery and they discovered during surgery

that they had early stage cancer but
in practical sense we wouldn't need to do

that it would just take a blood sample
that would be ideal but are we there

the.

Right.

Yeah I don't know the answer to that but
you know I can answer something else.

We've actually done the same way that you
show you here for prostate cancer and

it works really nicely and we were very
lucky were the panel you know we have

a big success of identifying what
the molecules where and but you know so

we did that first work and we showed it to
a clinician an expert in prostate cancer

and the question was like you know this
is really interesting now we need to fine

tune it which is likely what you're saying
to see if we can predict outcomes and

the other problem is that it's unclear
many times they go through the first

biopsy and it's unclear if the patient
needs a subsequent biopsy sometimes

they do make multiple biopsies and
those layoffs is carry many.

Risks so Rockley working on that where
we're trying to predict what patients

need second and third biopsies based on
what we see in the first biopsies and

the national profile.

It's you know I agree with you that
prostate cancer I think it's more

manageable because there
are different pathways for

treatment you know it's
more standardized than

this.

Yeah most likely they are not so

many of these have to do
with inflammation OK So and.

So you can envision a scenario
where somebody may have.

So we know when we look at humans we
basically try to include that they versity

in the training said in an ideal
training said you would have

a population that has all sorts of issues
that could confound the results and

you would try to tease apart just me or my
parents are associated with ovarian cancer

I don't know the answer to that question
our said because he told fifty people so

they had other inflammatory processes
going on because they were going for

surgery but the control population also
had some benign O'Brian conditions OK So

that's the best thing that you can do to
ensure that you meet again that risk so

to do the same you experiment so
the normals are not completely healthy but

you also have some benign conditions that
may have some inflammatory process is

going on but you're absolutely right
are they specific we I don't know when to

say yes I mean we don't really know that.

Hopefully.

All right thank you.