What was the last five so
we had it was our

distinguished lecture was a joy to

my patients you know and the great life.

So I'm one of those that go.

The center of my life.

You know that little pleasure and

the order in the years I would just be so
rich or

nice as my life was ordered by the girls

in my service or
a symbol of our own British life.

They should rather sharp and the better.

THIS HOUR.

Let me just tell you that you can be
present by several just by this there

is just let the conviction of
your cross the threshold but

just let the process of your life and
there's a lot of it is also likely.

You're on generic cause that doesn't go by

this remarkable discovery
the subject of last night and

I really wasn't that little
bit by Michael Jackson

this out of the end he looked
like you better go make a fresh

original This is originally from work.

Years ago.
Yes you live in.

You know just like you know any other day
styles from Asian problem with George.

They're the most prominent right.

And that's what the issue of us got it

from the loss of your shirt from
his famous fantasy life and

six years from both I did read it and
greatest life and this is

my sixth day just don't read it.

Five years old at that point and

I can see the change is this sort of

role model but I always get there and

you know you're standing right
over your head of your city.

Yes And then like that so was

it just kind of let what's there.

What's left of those
five years of good life.

You know I don't like this
discovery was made possible.

You're the liar.

You said that I want to go by
your methods of monies and

by night when I go with
you with my lips and

was just as it was just
like you said the people of

biologists while they're very
look to see what a science

has been top of it was
against somebody just

of a small measure to the support
of the leadership of.

This is where you can actually hear
this and it a good idea just to be

Hannibal Lechter for that matter you
should know better than the science and

what made us the robots at the scene.

Yes it was a bit of a guest.

This is a fight with your own.

Well first of all.

Michael published a number
of papers last year.

Always disliked it just a couple but

it's notices that he says
that many of us adults.

One of the biggest but
he was elected the number of years.

One was just

in the south of that over was
knighted by the News of the second

episode that made it possible
just standing there and

this on among many others
while it's not immediately.

Just bad but
is it all rich is it good music lovers.

And now he's one of the you can easily
get together with many social musicians

Stuart Andrew Lloyd Webber sort of
I'm joined by actually good at it as

well as Sir Paul McCartney This
is all my thing and not.

Richard Roth because it's
enormous on the sleeve and

the play well thank you very much.

It's always very embarrassing
to hear these introductions and

you shouldn't believe too much of it.

What I'm going to try to do
today is to give to talks

in one and the reason for that is that
we have a new initiative on the Y..

At the moment and I really want to solicit
some of your help on this initiative but

I also want to tell you what.

Restriction enzymes bioinformatics and

you know makes things that have been very
close to my heart for many many years.

So I'll start off by going quickly through
the you know mix of restriction of

modification and then move on to this
new project which is called calm breaks

which is all about annotating genomes.

So let me just remind you of what the
phenomenon of restriction of modification

is so bacteria have constant
problems because they're always

surrounded by bacteria phages viruses
that come and want to take them over.

If they possibly can.

And so bacteria have come
up with a kind of immune

system that will help them to prevent
pages from infecting them and

they do this by encoding an enzyme called
a restriction enzyme represented by this

little gremlin here with scissors and the
idea is that this gremlin will come along

this enzyme will cut every time it see
that particular sequence within the D.N.A.

in this case G T T C is the recognition
sequence for Echo or one.

This is one of the well known enzymes and

the idea is that as the pages
injecting its D.N.A. into the cell

the restriction enzyme will cut it up into
pieces and so will stop the infection.

Now of course this could be a problem for
the bacterium if it didn't have some way

of protecting its own D.N.A. against the
action of the restriction enzyme and so

it has another enzyme represented by
this gremlin which actually modifies

this specific sequence that is recognized
by the car one restriction enzyme

it puts a metal group right on
the second adenosine residue and

in so doing this protects the host D.N.A.
the.

Bacterial D.N.A. against the action
of the restriction and so on.

And the bottom line is that you end up
with a competition when a new phase D.N.A.

is entering the so
between the modification enzyme

which is in the cell and has been keeping
the host cell D.N.A. protected and

the restriction enzyme which is looking
for some unmet fellate the D.N.I.

to cut a nature usually has stacked
the deck so that what happens

is that the math the laser is not a very
good enzyme it's just not very fast and

that's where is the restriction enzymes
tend to be very very fast cutting and

so most of the time the restriction
enzymes gets to the site before it can be

methylated cut set up and the cell is
protected and this is very good for

a number of reasons but
most importantly it's very good because

in the laboratory these restrictions and
zines work very quickly.

They're very efficient.

They work for us.

Typically you can get complete digestions
in a few minutes with them and

this makes it very easy to
do biochemistry with them.

So there's a lot of these
things are now known and

I'll tell you about some of these i just
want to give a plug to my database called

replace see if you google ribeye if you
can easily find it and this is basically

a database of everything that you might
want to know about restriction enzymes.

It's actually one of the very earliest of
the molecular biology databases this was

started in one nine
hundred seventy five and

has been continuously
maintained since then.

Now let's talk about the restriction
modification systems because

the enzymes have been so
useful in the molecular biology lab and

for the whole genetic
engineering industry.

What's happened is that there has
been over the years a tremendous.

Enzyme and as a result of this.

Which is actually very easy to carry
out all you do is grow a bacterium

make a crude extract throw in some D.N.A.
and then ask.

Can the extract cut the D.N.A.
into nice discrete fragments and

you can display these on an eye gross gel
very easy to find new restriction enzymes.

And as a result of that most
of the enzymes shown up here

type two have actually been isolated and
shown to be functional.

So there are some more than thirty
eight hundred of these and so I'm sorry

been discovered during the course of
screening both in academic labs but

also in commercial companies
including New England by a labs so

there are very large number
of these an unknown and

they've been characterized in fact among
the enzymes that have been characterized

biochemically this is probably
the largest single family many more of

these have been found in any other enzyme
you can think about and they've actually

been characterized in the laboratory in
terms of their recognition sequences.

But we now know that there are actually
a lot of different types of restriction

enzymes we have four major types
type Rawn enzymes that will

recognize a specific sequence but they
come randomly away from the sequence and

they have some quite interesting bio
physical properties but they turn out

not to be useful as reagents to cut up
the and I because of this random cleavage.

There are three subunit enzymes and

I don't propose to go into them in any
detail other than to say that we have

evidence for ninety four functional
enzymes that is either genetic or

biochemical evidence showing
the active and that they would.

The type two of the ones that I
just talked about but you have two

completely separate genes one code for
methylation one code restriction enzyme.

Type three enzymes also have two genes but
in order to get restriction.

You have a common.

Nation of these two subunits the Reds and
the mob subunits and

just fourteen of these have been shown to
be functional and finally there's a class

here called Type four and
these enzymes recognize methylated D.N.A.

Now you have to realize that
bacteria in code restriction enzymes

are busy fighting for aging and so
of course the for a huge fight back.

And so what would you do if you
were afraid well we figured

let's modify our D.N.A. So the restriction
enzymes won't recognize us anymore.

And so in fact quite a lot of traders
that modify D.N.A. by methylation or

in other ways and in this way they
overcome restriction barriers and

as a result of that the bacteria
then figure out well

we've got to do something about these and
so they developed enzymes that

would specifically recognize
modified D.N.A. And so

that now a whole bunch of enzymes that we
know that recognize methylated D.N.A. and

there is a little bit of a man creature
issue here because some of these and

zines have been classified as type two and
some this type for and that's something

that I'm going to be doing something
about over the course of the next year.

So the nomenklatura becomes
a little more consistent.

Now you'll notice over here.

I have a column called putative Now
it turns out that rather a large

number of these and signs we've determined
we've cloned and we sequence and

as a result of those sequences
we can now go into Gen Bank and

into the sequence microbial genomes.

And we can find examples of genes that
we expect to be coding for restriction

enzymes in the case of the type one
systems some two thousand putative systems

are found among sequences that are in
general by this is among about twelve or

thirteen hundred prokaryotic sequences
engine bank and then just other sequences

that we've gone in for other reasons
in the case of the type two enzymes.

There are.

Some thirty six hundred shows some
similarity to type two enzymes type

two restriction enzymes or we can guess
on the basis of structure protection but

they should be restriction enzymes.

But there's more than eight thousand
D.N.A. methylation genes that we can find

and I'll tell you why there is
this discrepancy in the numbers

in a little while but you can see
this is a pretty large number and

it greatly exceeds the numbers that have
been functionally characterized and this

is an important problem that I will get
back to in the second part of the talk.

In this case we have six hundred type
three S. that we can recognize and

here we've got some fourteen hundred
thought forms that we can recognize but

it's really these types of foods that
have consumed most interest because these

are the ones that make for
good religions for bio chemistry and

you can imagine with some
three thousand represent more

than three thousand representatives
they come in a variety of flavors.

So there are some that are just the simple
two genes a restriction enzymes gene and

the methylated gene.

Some have an additional gene with them
which we call a controller gene and

this C. gene controls the expression
of the restriction enzymes gene so

that if the whole system is going to move
from one bacterium to another as these

systems often do you can imagine if you
move both genes into a naive host and

express the methylation expressed
the restriction enzyme because a new

host will have thousands of sites
typically for the restriction and so on.

You just can't protect it in time and so
what this controller gene does is holding

check the expression of the restriction
enzymes gene until such time as

the new host is completely methylated I
could tell you a lot more about that but

that would be a whole seminar all in
itself very interesting set of genes here.

Now so many times recognize
asymmetric sequences if you

notice these two sequences
are symmetric if you write the sequence

five to three on one strand five to three
on the other it's exactly the same.

And so that means a method that
would recognize say this a residue

will also recognise the a residue
on the other strand.

But in this case where you have
an asymmetric recognition sequence.

You have to protect both strands to
protect the D.N.A. after replication and

in this case you find two genes two
methylation genes next to one another.

Occasionally these refused and so

this becomes very nice when you're
trying to analyze the genome sequence

because you can tell when you're looking
at an enzyme that is likely to be

recognized in an asymmetric sequence
because it comes with two metal is genes.

Sometimes the restriction enzymes gene
comes in two parts as in this case here.

And again one of these units will
come one strand sequence and

one will cut the other
strand of the sequence

is an example of a system that
inherently looks like a type one system.

It's got both M. S. and R.

subhuman it's also got one of
these controlling sequences.

But it recognizes a symmetric
sequence one of the more interesting

classes is one of these are M.
systems in which both restriction and

modification are encoded
in exactly the same gene.

This is what we call the M M E one family.

There are something like two to three
hundred of these that we know about

at the moment and they're particularly
interesting because they're one group of

restriction enzymes where not only
have we been able to work out how

the recognition takes place and
then to alter it by genetic engineering so

that we can now make a large number of
enzymes that recognise sequences of

general asymmetric structure but
we've also.

When able to look at music once
there's a make a sensible guess about

what the recognition sequence is going to
be and that turns out to be useful to.

Finally die and here we have a case
of an enzyme H.P.A. to that has this

extra gene it's called a v gene and
this is a part of a repair system.

So now it turns out that this method
transfer is forms five methyl C. and

five natural C. is inherently mutagenic
and the problem with that is that

if you're going to be constantly mutant
uniting what happens five methyl see just

the emanates very easily a mix T.
and if you didn't do something about that.

Then after replication you would get
a mutation and it turns out that there's

very gene is a mismatch recognizing
enzyme that recognizes when you have a T.

G. mismatch specifically within
the context of this recognition sequence

will cut their teeth containing strand and
thereby open up the mutant

in order to get it repaired and so
this restriction system realizing that

it's inherently causing a problem brings
along some repair machinery to go with it.

Now there are three kinds of methylation
that we know about protecting zines

protect D.N.A. against restriction enzymes
five metal sites a scene that I just

mentioned in which the metal group is here
at the five position on the aromatic and

then two extra extra bring
their external methyl groups.

So in four metal cytosine and

six metal Adney in which this metal group
is on this extra X. so cyclic aiming

struck up here these turned out to be very
easy to make it's an easy character or

chemical reaction to carry out to put
a metal group onto an external A mean.

Whereas this is a very
difficult thing to carry at.

These three groups all have a common.

If you look at the hands
arms that produce and

by all have some similarities in
terms of the sequence of the enzymes.

But it turns out that enzymes that do
this function a very easy to recognize by

bioinformatics But once the this or
a little more difficult and

in particular we still don't know how
to differentiate just by looking at

sequences whether an enzyme is going
to be an end for methyl C N sign or

an end six methyl a enzyme So
I want now to talk to you about how

we go about analyzing genomes in order
to find new restriction systems.

So the any genes a really easy to
find because they have good sequence

multi-faith within them and you can think
about this in terms of the fact that

they do two things these enzymes
first of all recognize D.N.A.

but they recognize a lot of
different D.N.A. sequences and

so they will have regions within them that
are very variable because they have to do

with sequence recognition but they also
have a section of the gene protein

that is necessary to do the chemistry and
these are concerned this is the same.

And so
that is makes them pretty easy to find.

Now these as some of the genes
the acid the specificity subunits of

the type one enzyme and the V. genes
are the genes that do the mismatch repair

these are pretty easy to find and so
when you see one of these you can think.

Well yes we must be part
of a restriction system.

Do you see genes the genes that control
expression of restriction enzymes.

Some are quite easy because
there's one big family of

these things the we've located.

But then some are quite difficult.

There are a few that look as though
they've evolved in some separate way.

And finally the things we're
really interested in the our genes

the restriction enzymes genes are very
difficult to find and the reason

that they're very difficult to find so
I skipped that slide the reason that the.

To find is that they're
evolving very rapidly and

it turns out that even when you have two
restriction enzymes genes that recognise

exactly the same sequence very often
they show very limited sequence

similarity in some cases to a point where
you know you would never guess that they

were actually recognising the same
sequence and this is because that part of

host parasite interaction constantly
being challenged by you for

ages and also they don't have any direct

selection pressure on them except
when there is a challenge by afraid.

And if you happen to be in a situation
where you're in an organism with many

different restrictions systems and I'll
show you an example of rat in a moment

provided one of the restrictions
systems is active and

can dispose of the phage coming in.

Then there is no pressure on the other
systems to maintain their activity.

And this is reflected by the fact that
they're changing extremely rapidly.

Now if we go through and

look at the sequence microbial genomes as
of yesterday there were twelve hundred and

eighty five bacterial genomes for which we
have complete sequences and I mean closed.

Sequences not just shotgun data we
have ninety three are calle sequences.

And among all of the eleven
hundred fifty five of them have at

least one restriction system in them.

The ones that don't tend to be
bacterial endosymbionts things that

live inside host cells
many examples of this.

They apparently don't face problems for
from Fader's or

if they do they find out of
ways of dealing with them.

Now a lot of what we do is to go.

Sorry I went the wrong way that
a lot of what we do is to look at

the newly sequenced genomes and see what
we can find by way of restriction says.

Thems and a number of years ago
Helicobacter sorry Helicobacter

pylori was sequenced and it was discovered
that in this organism there are no

less than twenty six different
restriction systems present and

for some reason this organism is just
taking up restriction systems and

what I show here is not all twenty six but
just type.

Two because these are ones
in which we actually

cloned every single example
of these systems and

tracked Frank tippity and what we
found was that among these systems.

There were six that had active
prescription enzymes associated with them

but were eleven that had active
methylated genes associated with them and

the rest of the ones that are active
are the ones that are labeled

H P Y A five H P Y A three and so on.

They have a peer after them.

If that putative that is if they were
just identified by Bioinformatics.

But now a lot of Helicobacter pylori
strange different strains that have been

sequenced.

And so far no one has found
a strain in which the two strains

have exactly the same complement of
restriction enzymes being expressed and

methane transferees is being expressed.

And so it may be that in some way
these provide an epi genetic mark

in the case of bacteria to somehow
prevent somehow identify the species and

prevent cross cross breeding of any sort.

So this is rather an extreme example it's
not the most extreme example but there

are a lot of organisms that would have
four five six restriction systems in them.

So this gets to be quite interesting.

It also makes life a little more
complicated and it also explains.

Why these enzymes are evolving so
rapidly because in a case like this you

know if this enzyme is sufficient
to actually blog entry by the face.

Then there's no selective
pressure on the others and

it's only when a freight comes in.

That is modified against all one
of the systems that you then

have to worry about that
system being active.

It also turns out that a lot of these
systems that have just active methylation

genes have right next to them
a restriction enzymes gene but

is one mutation away from being active and
in two cases we've shown that

directly that you can reactivate
by a single mutation.

OK So there's a slide that was missing.

I just put it in the wrong order.

So I've already talked
about that we can press on.

So now what I want to do is
to show you the results of

an idea that I had while I was
taking a shower one day one of

the things that I think is kind
of fun about bioinformatics

is that there are all sorts of ways in
which bioinformatics can do things that

are a lot more interesting just
looking at sequence comparisons and

it occurred to me one day as I said
I was taking a shower at the time but

when you do a shutdown
sequence experiments so

you know I clone a bacterial genome I
do shotgun sequencing of everything.

Typically you clone two to three killer
bass fragments you take five hundred

base pairs reeds from each A And so
the question is what happens to fragments.

But contain intact non-clinical genes
such as restriction enzymes genes.

So you know let's say one of these two
K.B. pieces had a restriction enzymes gene

and it's well when that goes into a colon
and expresses it will kill the colon.

And so any such genes should be
missing from the set of clones.

But to find the shotgun.

That's illustrated on the next slide where
you can see that if you've got a clone

that runs up here just goes a little way
into the restriction enzymes gene then it

should be perfectly OK there's no reason
to think it's going to have a problem but

as soon as you start to take
clones that start over here and

run into the restriction enzymes gene and
contain the complete gene they're

going to be dead and they're going to be
missing from the shotgun sequence set.

And eventually you'll get to a point where
perhaps you can clone something from

the middle of the restriction enzymes gene
and now it will proceed perfectly well and

of course all the ones
down here will be OK too.

And so we thought well I thought what we
could do is we could look at all the shock

and sequence data and see where
there were gaps in the coverage and

in principle those gaps should
correspond to lethal genes and

in particular in this case
the restriction enzymes genes and

they should tell us when
restriction systems were active.

Unfortunately there was plenty of data for
this marvelous

influenza was the very first genome ever
sequenced it was done by Craig Venter

they had the original data they had to go
down in the basement and gather it got me

the original data and we started looking
and what I've done here is just to plot.

The position of the five crime end of
the READ going from left to right.

And these are the two genes down here
here's the restriction enzymes Gene Here's

the methylated gene and

what you can see is that upstream
of the restriction enzymes gene.

There is a big gap because any clones
starting in that area would have

gone straight through the restriction
enzymes she would have had it intact and

would be dead and if we do
the same thing on the other strand

we see here that there is also
a gap upstream of the restrict.

An enzyme Gene except for
this one guy here.

And when we looked indeed what
was in that particular read it

turned out it was a crime.

Eric.
Clone that had a little bit of sequence

from here but the rest of
the sequence was from somewhere else.

Elsewhere in the genome.

It was just that when they
did the original shotgun.

Two pieces got fused together.

So we looked at a large number of
genes a large number of genomes where

the data was available and where we
knew what was active and what wasn't.

And it turned out this was a very
nice way of finding active genes.

So the next thing we did was to go
to a strange way we never looked

before where no one had ever done the
biochemistry and we looked at the math and

a carcass no metal a carcass in
this case a metal Akaka strain and

bioinformatics we could see there
were three nice genes up here there's

a methylated gene this at the time
was just an open reading frame that

didn't look like anything else in German
bank and then down here was one of these

very genes the thing that is responsible
for mismatch repair and when we look at

the reeds you can see down here there
is a nice gap going from left to right.

Upstream of the restriction
enzymes gene and

on the other strand there is a nice gap.

Upstream of the restriction
enzymes gene in the other way so

we were pretty certain this was going to
be an active restriction enzymes gene.

So we clone did that and we tested it.

This was just we actually tested it
in vitro transcription translation we

just amplified a little bit of D.N.A.
that contained the gene put it

into an in vitro transcription translation
system added some land to D.N.A.

and ran it out on a gel and you can see
here you get a nice banding pattern.

This one is actually
the results of taking just

this in vitro transcription
system from the caucus gene over

here we recognize this pattern actually
And so over here is the pattern you.

Get from an enzyme called B S S H
to a known restriction enzymes.

And in the middle is what happens if you
mix the two just to show that these two

banding patterns really are the same.

And if they were different.

You would get extra bands in the middle.

When we looked to the side of cleavage
we discovered that whereas B.

S.S.H. two left for prime four base

three prime overhang this one left
the two base three prime over.

Sorry the five prime overhang
to base three prime overhang.

So the two ends are to have no sequence
similarity they don't look anything like

one another and
in fact we found a number of

other you restriction enzymes
doing exactly the same technique.

So I thought this was kind of
a funny way of using bioinformatics

to to do things that were not
immediately obvious and yet

actually you get functional
information out of this kind of data.

Now unfortunately everybody's
discovered always hard throughput

sequencing in cloning any more.

And so the amount of data available for
this is gong.

So we can't keep doing this for
a while it was pretty good.

So now what I want to do is to
talk to you about a new project

is just getting started and
it's called com breaks it's

called computational bridges
to experimentation and

it sort of builds on something that has
been a focus in my lab for a long time and

that is I like to do bioinformatics but
I like to sort of make predictions and

then test them I like to link
the bioinformatics to the biology and

I have an experimental lab in addition
to several points from the Titians

who work with me back in two thousand and

four I wrote a paper a little commentary
that was published in past biology.

And it was basically saying that we have
a problem at the time there was a lot of

sequencing taking place we would
gathering D.N.A. sequences galore.

And what we were finding
was that there were high.

Hundreds and hundreds of genes that you
could identify using Mark's programs and

other people's programs you could clearly
see here is a nice coding region.

We have absolutely no idea
what those coding regions do.

And so my proposal because there are no
high throughput methods that we know

of to do this my proposal was that
we do a pseudo throughput method and

we achieve high throughput
by paralyzation and

the idea is you get all the bioinformatics
that you can who want to collaborate on

this to make predictions and put them into
a big database and so we have a great big

database of predictions and then we
recruit biochemists to come along and

test some of those predictions elected
ones things that are going to give you

a lot of value in terms of later
being able to do bioinformatics or

later being able to understand
the biology of the organism and

in this way we would
build a set of genes for

which we had well characterized
experimental functions for

instance it's already known there
may be among the genomes that we now

at the moment there are plenty of examples
of genes that occur in five hundred

six hundred of these organisms and
it's clearly it's the same gene.

We have no idea what they do.

Absolutely.

Unknown function and so this is a way of
trying to find out what the function is to

expand our knowledge so that perhaps
ultimately one day we really can do

systems biology because we know what
all the gene products are doing.

So I say this was proposed
in two thousand and

four I got a very positive response from
an I age they said let's organize a little

conference talk about this and
see how we can make progress go forward.

So I did that in late two thousand and

four I organized a conference down in
Washington all of the key people from and

I came to this conference of all people
associated with the funding agencies.

But the problem was that
what I wanted to do was.

To match up expert
biochemists with particular

predictions that lay right in their
area of expertise and in such a lab.

It doesn't actually take very much extra
money to support a student to come in.

It could be a rotation students could be
a graduate student could even be a high

school student in many cases but or
even an undergraduate to come in and

just sort of you know make a little bit of
the dream product put it through the ass

eyes that are well known in the lab and
find out whether this prediction is

correct or
not because at the moment what happens.

Loads of people are making predictions.

No one is testing them part of this you
know biochemist don't like to test other

people's predictions particularly.

So the idea was well what we'll do is
we'll give a small grant to the lab.

We'll keep them five ten thousand dollars
just in order to test this particular

prediction.

So that was the essence of this proposal
Well nothing happened until two thousand

and eight late in two thousand and eight
when and I each in their wisdom decided

I should as a result of this all appear
always funded said well you know you're

not asking for enough money we don't
know how to give away five thousand but

we know how to give away two
hundred fifty thousand dollars but

we don't know how to manage your
portfolio as small as this.

And so nothing happened.

I couldn't get them to do anything so
in two thousand and eight.

Finally an R.F.P. request for proposals
came at tackling exactly this problem.

How are we going to do annotation
better because even and

I finally realized in
light of the numbers.

You will see here this is the growth in
the number of microbial genomes coming

at this is two thousand and four and
here when I originally wrote my proposal.

They're getting worried they're
spending all this money on sequencing.

But the functional annotation of
those sequences is lagging way

way behind and you could at this
point I think make a strong.

A swell was a point in sequencing anymore
genomes if you're not going to try and

work out what the D.N.A. is doing.

Why bother gathering more D.N.A. sequence.

Well you know obviously D.N.A. sequences
don't find that a very popular idea.

And so they felt I had to do something
about this functional annotation problem.

So I put out this request for

proposals we were applied to it a group
when I say we aren't talking about two

Boston University where I have an adjunct
position and I and wrote a proposal that

essentially flashed at this original idea
that I've had in the plus biology article.

I just want to illustrate the problem and
it really is significant.

So mufflers influenza was
the very first genome

ever sequence this was done back
in one thousand nine hundred five.

And if we look at the content
of the genes you discover that

of the sixteen hundred fifty seven
genes that have been identified.

There are twelve hundred twenty eight that
are either now because someone's actually

done some experiment on them or
you can make a good prediction.

Now what do I mean by a good prediction
multiple clear I mean a prediction that is

based upon some computer looking and

saying here's a gene is ninety
percent similar to this gene and

another organism which has been
annotated in some particular way

the hydrogenated it's an oxidizer whatever
it is and the substrate is known.

However there is a problem.

So what happens very often with
computational predictions is

that you've got a gene that may or may
not be what I call a gold standard Gene

as you will see in a moment where some
point someone actually did an experiment.

And you find another gene that
is ninety percent similar to it.

So you say this must be the same and so

it gets annotated in German bankers
being the same and at some point.

Things go wrong.

This gene that was originally
ninety percent similar

to the gold standard gene one
where the function was done.

Somebody finds a gene that is
ninety percent similar to that and

say it must have the same function too.

And so what happens is the quality of
annotation tends to go down because

almost all of the annotation is being
done by computer programs not by people

computer programs are only as smart as
you tell and they should be very often

things that ninety eight percent identical
can actually have different substrates.

Many examples of glucose or
hydrolysis things that cut up sugars.

But you missed a couple but
you know acids difference and

all of a sudden the sugar that it
will cut next to it has changed.

Instead of recognizing right.

Recognizes silos and
so on and most computer

programs that are just doing all of this
annotation these days don't recognize that

you know ninety eight percent similar
wonderful it must be the same.

And so there is a lot of wrong and
attention in G.M.

bank some people best are made and
there might be as much as fifteen

percent of all of the annotation
in Jenin bank is wrong.

You cannot be trusted.

The bottom line is you
really cannot trust it.

This point since we don't know
which fifteen percent that is.

So evil these twelve hundred twenty eight
genes where there are good put Asians.

There's a fair chance two or
three hundred of these are wrong.

Also the we don't really
know what they do.

Now what's left is four
hundred twenty nine are known

these are genes that we
have no idea what they do.

We're told they're labelled as
being hypothetical protein.

There is some computer program of one
of Mark's programs came along and

said this is the start of the gene and
here is the end of the gene and

you look and sure enough makes
a perfectly good proteins for.

Fifty amino acids long looks
like a genuine protein.

We have no idea what it does it could
be similar to something else in twenty

other genomes probably don't know what
it does in any of those genomes either.

And so that is a bit of a problem and

then there are others that
are sort of hypothetical concern.

So that's what this
twenty six percent is and

this number is actually very close to what
it was back in one thousand nine hundred

five when the genome was first sequence.

There's not been a lot of progress
in understanding what the genes

are coding for.

In some of these other organisms
the numbers are higher

than a carcass was the first our kale
genome done thirty five percent.

No idea what the genes do.

One of the more recent ones
is Pseudomonas aeruginosa

where we have forty two
percent of the genes.

Almost half of the whole organisms genome.

We do not know what it is no idea
what it is doing and yet for

many of these genomes people are trying
to do what I call systems biology.

They're trying to work hand computer
programs that will tell you

how these genes are working
how the organism is working.

I think it's not doable.

The moment that is just simply not doable
but there's no reason to think that it

couldn't be doable and I think we should
focus on at least a few of these organisms

working out exactly what all the genes
are doing and that means doing two things.

It means.
First of all working at biochemically

that the genes are capable of and
then ideally genetically

putting it in within the context of
the working of the whole organism.

But the thing that is easy
to do is the biochemistry

the genetics Often all it ever
does is to give you a hint as to

what's going on maybe give you
a prediction as to what this gene is doing

it's not often that the genetics
absolutely nails the function

biochemistry doesn't a over function
it nails the in vitro function doesn't.

Tell you what pathways it may
be participating in and for

that you ultimately will
need to do with genetics.

But very often just knowing the
biochemical function can take you an awful

long way towards understanding
its role within the cell.

Sorry wrong way again.

So here's the cause of the problem you
know you hear a lot about how great it is

to be able to sequence D.N.A.
for less and less and less.

This is the cost per base pair.

The number of base pairs you get per
dollar and you can see what's happening

that the number of base pairs you
get per dollar is just going up and

up and up is soon going to
be right up to the sky.

It's amazing how cheap it is now
to get D.N.A. sequence and so

of course you know it's simple to do so
everybody does it but I ask myself what

is the point I ask you what is the point
in gathering all of this sequence.

If we're not going to try and work out
what it's doing if we're not trying to

try to understand the meaning
the biological meaning less

so why has there been so little progress.

Well there's a bunch of interesting
reasons one is the inherent difficulty

of the problem if I give you a protein of
three hundred fifty amino acids long and

I say hey I'll give you
a hundred thousand dollars work.

What it's doing.

You can't do it all you
can do is sort of Asif

everything that you think it might be
doing if you're lucky you'll hit it.

The odds are you're not
going to hit it and

so we definitely need to find ways of
making reasonable predictions about what

it's doing does it look as though it might
be interacting with accidental some of

finding does it look as though it's going
to need A.T.P. does it have motifs in it

suggests that it might be involved in this
and this is where the bio information is

can come in and there are many ways in
addition to just looking for motifs or

just looking for a sequence similarity
that you can make predictions I showed you

one of them in the case of the restriction
enzymes genes where you could act.

We identified genes that were lethal from
data that was carried in some completely

other way and there are many other ways
you can perhaps identify operands and

say Here's a bunch of genes five
genes that clearly all involved

in human bio sentences but there's
one open reading frame here right in

the middle of this opera and
we don't know what it does.

Can we perhaps make a good guess
as to what it might doing.

We could guess it's involved in here
by a sentence or two in some way.

And so
maybe there's a biochemical step involved.

For which we don't have a gene
you can go and test it.

So there are many ways in
which you can do this but

it typically takes a combination of
biochemistry maybe some genetics.

Maybe some computation to begin to
get a handle on this problem and

not a lot of people
have all of the skills.

So this cross disciplinary set of skills.

That's needed to really make progress
does not reside in very many labs

a lot of labs just focus on computation
others just focus on biochemistry and

a lot of them never talk to one another
and in fact a lot of chemist don't want to

talk to the bio information because they
think they're very upper to make it way

too many many predictions the other thing
that's happened recently is there's

this tremendous appeal for genome wide
studies everyone wants to do something

where you look at the whole genome all at
once you know let's not worry about these

individual things let's look at everything
and in fact it's becoming increasingly

difficult to get grants in areas where
genome wide studies are popular.

If all you want to do is actually look at
a few genes study sections say why don't

you look at the whole genome Why don't you
do micro rays Why don't you do this other.

Well unfortunately there are no genome
wide ways other than computationally

of getting this process this
problem to function and

then finally there's a lack of appropriate
funding mechanisms you cannot go to and

I can say hey I found an interesting gene
here it's present in five hundred genomes.

I want.
Work out what it does no one will give you

money for that you simply can't
get the money to do that.

So this is where I come bricks comes in

through put by paralyzation so
we get the bioinformatics sions on board

to make high quality predictions
things that really say.

I think this is an oxidizer and I think
the substrate is likely to be a sugar or

I think the substrate is likely to be
a turkey and I think it's likely to be

an alley frats it can mean an hourly
phatic substance of some description so

basically we want to get some high quality
predictions we want to assemble a database

of those predictions and we're in
the process of doing that at the moment.

Biochemists can then come and
test the predictions and initially we will

approach biochemists when we have a good
set of predictions where on the Web site

assembling a list of what we think are the
top one hundred per predictions things

that if these predictions turn out
to be accurate and can be tested and

shown to be accurate then that will
be very helpful to the information

in making better predictions later on and
so that is something the an important

criteria here because you know you write
an algorithm you make predictions.

If old predictions are correct you
think are the algorithm is great but

if half of them are wrong then you've got
to go back and readjust your algorithm so

that you can get the right answer.

More often.

And the idea is that by
getting biochemists and

having them work in their field of
expertise the incremental cost of doing

this work is really very very slight.

And so even though it's quite sly you
still have to persuade them to participate

so we offer them some financial
incentives to do this

given five ten thousand dollars and
the way one can think about this going.

Is that maybe there's a student
here at Georgia Tech maybe one

of you in the audience who says.

I wouldn't mind doing this as a rotation.

During my first year as a graduate
student and so I will go and

look in COM bricks and I'll see where is
an interesting gene something we don't

know the function for there's a very
good prediction you look you find it you

say all is a professor at Georgia Tech
who is actually working in this area.

And so between as we can apply to
combat for a grant perhaps get five or

ten thousand dollars so that I can go and
test the prediction in this person's lab.

So this is very good for
the lab to get a little extra money.

Very good for the student because if in
fact they can validate that prediction.

You can publish that result will be
a publishable result and in the process

you will help both the bioinformatics and
the biochemical communities so

you do a very valuable function
a very valuable service for

biology as a whole and the nice thing
about this is that there are so many of

these genes I mean there are thousands
upon thousands of these genes but

no one has to do exactly the same training
everyone can do something different.

You can all make a contribution and
it can be really very valuable and so

we're hoping to get a lot of momentum
within the community to do this

a lot a lot of opportunities
a lot of possibilities.

So one question that comes up had can
we improve the predictions Well I think

one of the things is we should recognize
the vast bulk of the current annotation

is really based on sequence similarity
to previously annotated genes.

This is OK but there are some problems
with it that I will point out in a moment.

We have to recognize a lot of
the annotation is wrong for

exactly this reason I quoted earlier that

something is ninety percent some similar
something we know the function of and

then something is ninety percent
similar to that and ninety eight.

Scent similar to that before you know it
you're ten percent similar to the original

gene and
the chances of it being correct a zero so.

We have to do provide a much better
basis for similarity based annotation.

That will be the subject of
the next slide and then we

will provide within the complex web site
a means for reporting Messina tensions.

What.
All of this boils down to is that

we need a gold standard set of genes and
proteins we need to

have this set where we can say we
know the sequence of this protein.

We know the function of this protein
because the biochemistry was carried

out on this protein with this sequence and
then have a standard we have

a gold standard against which we
compare every new gene that comes into

the database we can say how far away
is it from the gold standard and

at some point it's sufficiently far away
you say well maybe we better just recheck

make sure that it still has the same
function it's eighty percent

similar maybe we should check maybe
it's not really got the same function.

So the question is Where would
you go with such a gold standard

set where you might think you know
we've got big data massive databases

we've got gem bang we've got the D.N.A.
database in Japan.

We've got you know prop which is
supposedly the database for proteins.

We have the whole I Database Gene
both of which have huge

lists call I.D.'s and functions
associated with them and you might think.

But these would already be gold
standard databases they're not.

If you go to any of
the right databases and

you can we find out for
this gene where this function

was determined was in fact
the gene sequence upon which

this function was determined we discover
you can't even find that information.

This is not something that needs databases
of considered worth tracking and

what's happened in the.

Color is that perhaps someone in
some in color strange determined

are this is the D.N.A.
methylation D.N.A. methylation E. coli.

But perhaps the function
was actually tested

in a strain that was a long way away
from anything that's been sequenced and

in fact the gene was never
sequence to the time.

And so on and you discovered a large
number of the whole of proteins

fall into this category the sequence
that is in the databases this

ecology in may not be the one where
the function was determined and

so and the same is true you know crowd and
in fact

I can tell you all of these databases
are rather embarrassed by this fact

because when I asked them this question
everybody said no we can't provide that.

And when I proposed this idea for

a gold standard database
everybody said What a great idea.

We should get this.

And so now they're all busy
cooperating with me and

with Combray some with us to
actually assemble this data set.

So this is the essence of it but
each protein in the set.

We're going to find a reference that
describes the determination of function

going to make sure that the strain
is well defined and noted.

And we're going to make sure especially
that the protein sequence is known for

this protein in which the function
was determined if it doesn't for

if we can find this match
between a protein sequence.

It will not go into the gold standard set.

So we're putting this together it will
probably take a couple of months to

get a reasonable start on this because
both N.C.B.I. And you know products

are busy working out procedures to make
this easy but we have got started on this.

We have about a couple of
thousand candidate genes and for

each of these genes that being sent out
to manual curators who would go along.

Through literature and

I actually find whether or not
the literature supports the assignment.

So we identify candidate side from
databases or from individuals and

we welcome anybody who wants
to submit a gold standard

what they think should be a gold
standard on the complex web site.

You know prot will then make sure that
there's a unit product number for

this they will go through their annotation
and prepare a template that in principle

contains reference to the sequence
reference to the function and a few other

things about the strain designation and
so on and then these are being sent

to curators this is something
we're doing as part of the complex

project at the moment but ultimately
will be taken over by and C.B.I..

And what we're asking is just the curators
manually check the strain information

is accurate and sometimes people say well
you know we did this on the color line.

Well that becomes unknown strain.

It's OK that's fine but
one should not therefore assume that it's

the same as something that is
in the sequence databases.

We want to make sure that the biochemical
characterization is accurate

description that is in unit really
matches what was done and then

the fact that the gene was sequenced and
hence the protein sequence is known and

from exactly the strain in which the
biochemical characterization was done and

I just put a note here you've
got to know you know the sort of

the gold standard sequence we call lions
for a strain M G six hundred fifty five.

There has not been very much
characterization done on genes

from the strain so
it's always important to know where it was

to make sure the sequence is
the star the same as M G.

Sixteen fifty five before the annotation
is put in as gold standard.

So how can you help.

Well there are many ways in which
people can help you can send

in candidate gold standard genes and

proteins you can volunteer to serve
as a manual curator you just.

Rich Roberts Roberts said anybody dot com
and you can browse the complex web sites

and let us know if you see genes should
be gold standard the complex website

is complex you got edu again if you
google complex you will find this.

Now how are you going to prioritize
Gene since I told you you know

there are hundreds of
thousands of these things.

Well the first thing we want to know how
many organisms contain this gene for

which there is a prediction and
the larger the number of genes and

especially if it transcends
kingdoms the implants if it's in

animals then that gets to
be a lot more interesting.

So if any gene that is present a large
number of organisms Obviously if you

can work out what it does you
get a lot of bang for the buck.

We're also very interested in trying to
finish the functional annotation of E.

coli and so if the gene is
currently unknown in color and

if there is a good prediction and
that would be a high priority for testing.

Something for Helicobacter pylori we
wanted to show that this approach could be

used to annotate pathogenic
genomes obviously enter and

I aged very interested in pathogens.

Aleko back to Laura a very interesting.

This is the organism
the course has also but

it's also one of the few bacteria
absolutely known to cause cancer.

It causes stomach cancer.

It also causes this off.

Adriel cancer.

It can be grown in the lab and it's
actually relatively easy to work with.

Is there a clone or purified protein
available the structure of you

know makes initiative has actually
generated many thousands clones

of interesting unknown open reading
frames that are present in G.M. by and

the idea there was that they would say
well let's make this protein Let's try and

crystallize that if we can get
a structure then perhaps we

can determine the function well it turns.

When you try that approach about
one in ten of all the proteins

you make actually crystallize and
then when you get the structure.

Most of the time you haven't got a clue
about function structure does not easily

give rise to function but all of these
clones and all of these proteins were

viable and we can provide those we
can tell you how to get those and

then there's a whole bunch of other
criteria I don't want to go into so

there's a whole list of things that make
it interesting for particular proteins.

Who can help.

Well the computational biologists and

the biochemist obviously
also geneticists can help.

Finally university students high
school students who are actually going

to run an experiment with our local high
school to have their students come in

to buy a labs and do some of these tests
as part of a high school fair project.

I was at the Intel Science Fair earlier
this year and I can tell you the kids who

were participating there from all over the
world are more than qualified to do these

kinds of tests if they go to a decent
lab and get some supervision and

I think you know you could easily go and
find high school students here

who could come and work in labs at Georgia
Tech that would be straightforward.

You know here professors there
are a lot of professors who retire and

go emeritus and
they love doing experiments.

Many of them take the opportunity
to go back to the lab.

Not looking at anybody in particular but I
think there are some opportunities here to

find finally of course we need
help from the funding agencies and

I've been talking not just
an eye to N.S.F. to him.

I have a Hughes for
the Welcome Trust in England a lot of

them of expressed interest because this
can actually be a wonderful teaching tool.

There are many ways in which you could
build this into undergraduate education

and many ways in which you can
run some of courses in which you

get students to participate
in something like this and

a lot of times you don't need
a lot of funding you know.

To get it going.

We're hoping to make this
into a worldwide effort.

It's an ideal project for developing
countries where they don't have a lot of

money for science but often there is a lab
that has a lot of expertise in an area and

can do this and
can use this to train students.

So I think there are lot of possibilities
for people to get involved in this.

And of course you can help.

These are the people that
are involved myself Simon to see if

who is a computational biochemist
than a Boston University

a guy who came out of computer science and
moved into Bioinformatics.

Martin Stephan is a bio chemistry and

he's looking after a lot of
the biochemical side of things.

Charles de Lisi is
an emeritus professor be you.

He was the guy who first
provided funds for

the human genome project
when it got started.

Daniel said Gray is a young assistant
professor at B.U. who looks at networks

he's interested in networks with
in bacteria and other organisms.

Stephen Saltzburg is a well known
computational biologist at the University

of Maryland.

Denis picked up again another well
known Byron from attention at Columbia.

All of our collaborators we already
have a lot of collaborators both on

the computational front and
on the biochemical front.

We have half a dozen projects already
funded to get going on this and

of course we're very grateful for
funding from the and

I ate they gave us a to get Grant on the
stimulus money so this money is going to

run ads and we're busy working out how to
get them to give us a lot more because

obviously we're just at the start of this
but I think the very exciting project.

It's something in which we can show that
we as biologists really can collaborate

because however you look at it.

Science is a collaborative
activity you know everything we do

builds upon stuff that's
been done elsewhere.

And I can tell you it is a lot more fun
to collaborate than it is to complete

a long standing collaboration with
my colleague here at the front.

Really and crystallographer we've worked
together for many many years and it's so

much more fun to collaborate than it
is to try to compete with people and

I hope that all of you will but
at least think about this and

see if there's some way in which
you can help and collaborate.

Thank you thank

you very much for
the inspiration to get the ball.

I'm going to buy over the place
in sequence is the book

to me which is increasing

all your problems and

you should see some problems.

This is your question my

letters for you with my

eyes help you need to construct to convert
double standards or anything else.

Even your heart.

I read well.

So I think first off let's
say you have a protein.

If you have a gold standard.

There are many ways in which you
can say how far away it is so

obviously just a simple
statistical test and

many amino acids are identical but
also you can look at motifs are key

motifs conserved in particular if you
know that within the gold standard.

There is a little area that forms
a catalytic sites all such substrate.

Condition site and

you can see that that is different
in this new gene there already.

You can begin to say well you know maybe
this new protein doesn't have the same

function maybe it's different.

Many of the programs that are annotating
genomes at the moment don't even

do something.

The simplest and so I think the idea
of providing the gold standard is that

it will really give the annotator.

The opportunity to do a much better job
of annotating things because they have

something to relate it all to and that's
not what is happening at the moment.

You know there are more of one

on one.

So we don't want to call it works

you call a cute little guy so

expression is actually a lot
easier than it used to be but

there are a lot of tricks and
I think one of the key things is you

really have to keep your eye out for
where people

have come up with good traits now one of
the things we're focusing on bacterial and

genes not doing any eukaryotic genes at
the moment in this particular project.

Partly because of expression but partly
because you carry out it genes are so damn

difficult and complicated even predicting
where they are can be a big problem.

I mean I'm sorry I discovered this split
gene thing because it just complicated

biology enormously.

But I figure if we can really get
everything going with bacteria or

an R. K. or and get a good group of
people together and get everything going.

We can at least show that it will work and
then we'll expand it into you carry out

later on so
that is the idea at the moment.

So call out in which we know that
you know it's nice that in message.

You know.

So we're not looking for
cellular function.

All we're trying to do in this particular
project is to get at the biochemical

function because if you want to go beyond
that you're talking about really a huge

amount more work and not something that
can be funded for a spall amount of money.

What we're hoping though is that what
will come out of this is that people will

find some new functions that can then lead
them to right are ones to explore it and

find out about the biology what that OK.

And what's right.

So for protein protein interactions I
think we would be satisfied at this point

if we could identify the binding partner.

So for
instance if it's a transcription factor.

What is the D.N.A.

sequence that it binds to if it's
part of a multi subunit complex.

What what is it a part of the right to
some of proteins you know that part of

a massive complex some of them
have enzymatic function but

a lot of them don't.

But just the fact that they're a part
of that complex is good enough for

our purposes at this stage.

So we're not trying to solve the whole
of biology we're trying to do something

pretty simple that will actually lead
to better annotation of genomes so

that's really what it's all about you

know

right.

So I told you we have this class of
enzymes that are recognized by represented

by M M E one this is a single train
restriction modification enzyme and

in this particular case we already
know how to do that we can make

many changes specifically in that
particular class of enzymes.

We have another class of enzymes that
we're just starting to make some progress

on where we think we will
be able to do the same.

Sort of thing and

that is our go buy a lab says not to just
keep offering more and more enzymes.

But to say you want an enzyme
with a particular recognitions

sequence specificity Telos and
we will make it so this is our goal but

most of the work up by a labs now
is our aim towards understanding

enough about specificity so
we can make new ones by design.

Yeah.

You don't.

OK Just a scratch I'm
sorry this is working for

first class restriction enzymes
structures from you know for sure.

Structures and watch.

It's a very mature for the woman so

that fall into about four
classes you can classify them as

four among the ones we know we don't have
a lot of structures we have right things.

Sixteen or eighteen structures at
the moment and we certainly are certainly

not fully representative of the range of
restriction enzymes that we know about and

in particular the thing that I would
like to see some structures for

examples of enzymes that recognise
exactly the same sequence and

cut it exactly the same point.

And yet show no sequence similarity and
we in a while you

have one such example of that although
it's not quite that Eco are one and

bam H one recognise different sequences
but the fall that almost super

imposable the structures and the only
three amino acids that they have in common

are the three that form the catalytic
triad that do the cleavage and

that's not of any use in terms of
predicting stuff which you are clearly

one among young people five

years we've got to see because they're
pretty much reacting in a fine.

Gesture to a late sense.

Right right.

And we have tried some of the.

At some cases you can make

predictions about what these
things are going to recognize.

But in general that is not that approach
has not worked terribly well with

the sequence differences
are too different.

So for instance I go I want Bama
each one which almost coincide.

If you try to thread one on
the other it doesn't work

the threading algorithms
don't get it the same.

So it's a problem structure protection in
general is a bit of a problem if things

are close and threading works very well
but as you get further and further away.

I mean lots of these have sort of you
know four percent similarity of props and

identity and that it's quite a problem.

There's that we be glad to send you

some if you want to try reading signals
the functional size trying to use them.

Why this mean.

Terrible shock if you want
to actually write but

the problem is that the catalytic triad in
Bam makes one an echo of one you've got

one here one here one here in
the sequence in space right space.

Right.

See that's the way it goes and so

you get at least in theory which might be

OK Well we'll send you some to try and
there was you one question I think.

Yes OK he's

right.

If

I was running and
I would do that but I'm not OK.

Yes.

Membrane proteins are a major.

Well I'm from every conceivable
function that you can imagine.

So from a structural perspective
from a functional perspective and

I think at the moment we're going to leave
the membrane proteins to the experts So

for instance some of them.

We know are transporters some are involved
in transporting things to and fro and

we really Milton Sawyer is
a collaborator of Oz

What was the last five so
we had it was our

distinguished lecture was a joy to

my patients you know and the great life.

So I'm one of those that go.

The center of my life.

You know that little pleasure and

the order in the years I would just be so
rich or

nice as my life was ordered by the girls

in my service or
a symbol of our own British life.

They should rather sharp and the better.

THIS HOUR.

Let me just tell you that you can be
present by several just by this there

is just let the conviction of
your cross the threshold but

just let the process of your life and
there's a lot of it is also likely.

You're on generic cause that doesn't go by

this remarkable discovery
the subject of last night and

I really wasn't that little
bit by Michael Jackson

this out of the end he looked
like you better go make a fresh

original This is originally from work.

Years ago.
Yes you live in.

You know just like you know any other day
styles from Asian problem with George.

They're the most prominent right.

And that's what the issue of us got it

from the loss of your shirt from
his famous fantasy life and

six years from both I did read it and
greatest life and this is

my sixth day just don't read it.

Five years old at that point and

I can see the change is this sort of

role model but I always get there and

you know you're standing right
over your head of your city.

Yes And then like that so was

it just kind of let what's there.

What's left of those
five years of good life.

You know I don't like this
discovery was made possible.

You're the liar.

You said that I want to go by
your methods of monies and

by night when I go with
you with my lips and

was just as it was just
like you said the people of

biologists while they're very
look to see what a science

has been top of it was
against somebody just

of a small measure to the support
of the leadership of.

This is where you can actually hear
this and it a good idea just to be

Hannibal Lechter for that matter you
should know better than the science and

what made us the robots at the scene.

Yes it was a bit of a guest.

This is a fight with your own.

Well first of all.

Michael published a number
of papers last year.

Always disliked it just a couple but

it's notices that he says
that many of us adults.

One of the biggest but
he was elected the number of years.

One was just

in the south of that over was
knighted by the News of the second

episode that made it possible
just standing there and

this on among many others
while it's not immediately.

Just bad but
is it all rich is it good music lovers.

And now he's one of the you can easily
get together with many social musicians

Stuart Andrew Lloyd Webber sort of
I'm joined by actually good at it as

well as Sir Paul McCartney This
is all my thing and not.

Richard Roth because it's
enormous on the sleeve and

the play well thank you very much.

It's always very embarrassing
to hear these introductions and

you shouldn't believe too much of it.

What I'm going to try to do
today is to give to talks

in one and the reason for that is that
we have a new initiative on the Y..

At the moment and I really want to solicit
some of your help on this initiative but

I also want to tell you what.

Restriction enzymes bioinformatics and

you know makes things that have been very
close to my heart for many many years.

So I'll start off by going quickly through
the you know mix of restriction of

modification and then move on to this
new project which is called calm breaks

which is all about annotating genomes.

So let me just remind you of what the
phenomenon of restriction of modification

is so bacteria have constant
problems because they're always

surrounded by bacteria phages viruses
that come and want to take them over.

If they possibly can.

And so bacteria have come
up with a kind of immune

system that will help them to prevent
pages from infecting them and

they do this by encoding an enzyme called
a restriction enzyme represented by this

little gremlin here with scissors and the
idea is that this gremlin will come along

this enzyme will cut every time it see
that particular sequence within the D.N.A.

in this case G T T C is the recognition
sequence for Echo or one.

This is one of the well known enzymes and

the idea is that as the pages
injecting its D.N.A. into the cell

the restriction enzyme will cut it up into
pieces and so will stop the infection.

Now of course this could be a problem for
the bacterium if it didn't have some way

of protecting its own D.N.A. against the
action of the restriction enzyme and so

it has another enzyme represented by
this gremlin which actually modifies

this specific sequence that is recognized
by the car one restriction enzyme

it puts a metal group right on
the second adenosine residue and

in so doing this protects the host D.N.A.
the.

Bacterial D.N.A. against the action
of the restriction and so on.

And the bottom line is that you end up
with a competition when a new phase D.N.A.

is entering the so
between the modification enzyme

which is in the cell and has been keeping
the host cell D.N.A. protected and

the restriction enzyme which is looking
for some unmet fellate the D.N.I.

to cut a nature usually has stacked
the deck so that what happens

is that the math the laser is not a very
good enzyme it's just not very fast and

that's where is the restriction enzymes
tend to be very very fast cutting and

so most of the time the restriction
enzymes gets to the site before it can be

methylated cut set up and the cell is
protected and this is very good for

a number of reasons but
most importantly it's very good because

in the laboratory these restrictions and
zines work very quickly.

They're very efficient.

They work for us.

Typically you can get complete digestions
in a few minutes with them and

this makes it very easy to
do biochemistry with them.

So there's a lot of these
things are now known and

I'll tell you about some of these i just
want to give a plug to my database called

replace see if you google ribeye if you
can easily find it and this is basically

a database of everything that you might
want to know about restriction enzymes.

It's actually one of the very earliest of
the molecular biology databases this was

started in one nine
hundred seventy five and

has been continuously
maintained since then.

Now let's talk about the restriction
modification systems because

the enzymes have been so
useful in the molecular biology lab and

for the whole genetic
engineering industry.

What's happened is that there has
been over the years a tremendous.

Enzyme and as a result of this.

Which is actually very easy to carry
out all you do is grow a bacterium

make a crude extract throw in some D.N.A.
and then ask.

Can the extract cut the D.N.A.
into nice discrete fragments and

you can display these on an eye gross gel
very easy to find new restriction enzymes.

And as a result of that most
of the enzymes shown up here

type two have actually been isolated and
shown to be functional.

So there are some more than thirty
eight hundred of these and so I'm sorry

been discovered during the course of
screening both in academic labs but

also in commercial companies
including New England by a labs so

there are very large number
of these an unknown and

they've been characterized in fact among
the enzymes that have been characterized

biochemically this is probably
the largest single family many more of

these have been found in any other enzyme
you can think about and they've actually

been characterized in the laboratory in
terms of their recognition sequences.

But we now know that there are actually
a lot of different types of restriction

enzymes we have four major types
type Rawn enzymes that will

recognize a specific sequence but they
come randomly away from the sequence and

they have some quite interesting bio
physical properties but they turn out

not to be useful as reagents to cut up
the and I because of this random cleavage.

There are three subunit enzymes and

I don't propose to go into them in any
detail other than to say that we have

evidence for ninety four functional
enzymes that is either genetic or

biochemical evidence showing
the active and that they would.

The type two of the ones that I
just talked about but you have two

completely separate genes one code for
methylation one code restriction enzyme.

Type three enzymes also have two genes but
in order to get restriction.

You have a common.

Nation of these two subunits the Reds and
the mob subunits and

just fourteen of these have been shown to
be functional and finally there's a class

here called Type four and
these enzymes recognize methylated D.N.A.

Now you have to realize that
bacteria in code restriction enzymes

are busy fighting for aging and so
of course the for a huge fight back.

And so what would you do if you
were afraid well we figured

let's modify our D.N.A. So the restriction
enzymes won't recognize us anymore.

And so in fact quite a lot of traders
that modify D.N.A. by methylation or

in other ways and in this way they
overcome restriction barriers and

as a result of that the bacteria
then figure out well

we've got to do something about these and
so they developed enzymes that

would specifically recognize
modified D.N.A. And so

that now a whole bunch of enzymes that we
know that recognize methylated D.N.A. and

there is a little bit of a man creature
issue here because some of these and

zines have been classified as type two and
some this type for and that's something

that I'm going to be doing something
about over the course of the next year.

So the nomenklatura becomes
a little more consistent.

Now you'll notice over here.

I have a column called putative Now
it turns out that rather a large

number of these and signs we've determined
we've cloned and we sequence and

as a result of those sequences
we can now go into Gen Bank and

into the sequence microbial genomes.

And we can find examples of genes that
we expect to be coding for restriction

enzymes in the case of the type one
systems some two thousand putative systems

are found among sequences that are in
general by this is among about twelve or

thirteen hundred prokaryotic sequences
engine bank and then just other sequences

that we've gone in for other reasons
in the case of the type two enzymes.

There are.

Some thirty six hundred shows some
similarity to type two enzymes type

two restriction enzymes or we can guess
on the basis of structure protection but

they should be restriction enzymes.

But there's more than eight thousand
D.N.A. methylation genes that we can find

and I'll tell you why there is
this discrepancy in the numbers

in a little while but you can see
this is a pretty large number and

it greatly exceeds the numbers that have
been functionally characterized and this

is an important problem that I will get
back to in the second part of the talk.

In this case we have six hundred type
three S. that we can recognize and

here we've got some fourteen hundred
thought forms that we can recognize but

it's really these types of foods that
have consumed most interest because these

are the ones that make for
good religions for bio chemistry and

you can imagine with some
three thousand represent more

than three thousand representatives
they come in a variety of flavors.

So there are some that are just the simple
two genes a restriction enzymes gene and

the methylated gene.

Some have an additional gene with them
which we call a controller gene and

this C. gene controls the expression
of the restriction enzymes gene so

that if the whole system is going to move
from one bacterium to another as these

systems often do you can imagine if you
move both genes into a naive host and

express the methylation expressed
the restriction enzyme because a new

host will have thousands of sites
typically for the restriction and so on.

You just can't protect it in time and so
what this controller gene does is holding

check the expression of the restriction
enzymes gene until such time as

the new host is completely methylated I
could tell you a lot more about that but

that would be a whole seminar all in
itself very interesting set of genes here.

Now so many times recognize
asymmetric sequences if you

notice these two sequences
are symmetric if you write the sequence

five to three on one strand five to three
on the other it's exactly the same.

And so that means a method that
would recognize say this a residue

will also recognise the a residue
on the other strand.

But in this case where you have
an asymmetric recognition sequence.

You have to protect both strands to
protect the D.N.A. after replication and

in this case you find two genes two
methylation genes next to one another.

Occasionally these refused and so

this becomes very nice when you're
trying to analyze the genome sequence

because you can tell when you're looking
at an enzyme that is likely to be

recognized in an asymmetric sequence
because it comes with two metal is genes.

Sometimes the restriction enzymes gene
comes in two parts as in this case here.

And again one of these units will
come one strand sequence and

one will cut the other
strand of the sequence

is an example of a system that
inherently looks like a type one system.

It's got both M. S. and R.

subhuman it's also got one of
these controlling sequences.

But it recognizes a symmetric
sequence one of the more interesting

classes is one of these are M.
systems in which both restriction and

modification are encoded
in exactly the same gene.

This is what we call the M M E one family.

There are something like two to three
hundred of these that we know about

at the moment and they're particularly
interesting because they're one group of

restriction enzymes where not only
have we been able to work out how

the recognition takes place and
then to alter it by genetic engineering so

that we can now make a large number of
enzymes that recognise sequences of

general asymmetric structure but
we've also.

When able to look at music once
there's a make a sensible guess about

what the recognition sequence is going to
be and that turns out to be useful to.

Finally die and here we have a case
of an enzyme H.P.A. to that has this

extra gene it's called a v gene and
this is a part of a repair system.

So now it turns out that this method
transfer is forms five methyl C. and

five natural C. is inherently mutagenic
and the problem with that is that

if you're going to be constantly mutant
uniting what happens five methyl see just

the emanates very easily a mix T.
and if you didn't do something about that.

Then after replication you would get
a mutation and it turns out that there's

very gene is a mismatch recognizing
enzyme that recognizes when you have a T.

G. mismatch specifically within
the context of this recognition sequence

will cut their teeth containing strand and
thereby open up the mutant

in order to get it repaired and so
this restriction system realizing that

it's inherently causing a problem brings
along some repair machinery to go with it.

Now there are three kinds of methylation
that we know about protecting zines

protect D.N.A. against restriction enzymes
five metal sites a scene that I just

mentioned in which the metal group is here
at the five position on the aromatic and

then two extra extra bring
their external methyl groups.

So in four metal cytosine and

six metal Adney in which this metal group
is on this extra X. so cyclic aiming

struck up here these turned out to be very
easy to make it's an easy character or

chemical reaction to carry out to put
a metal group onto an external A mean.

Whereas this is a very
difficult thing to carry at.

These three groups all have a common.

If you look at the hands
arms that produce and

by all have some similarities in
terms of the sequence of the enzymes.

But it turns out that enzymes that do
this function a very easy to recognize by

bioinformatics But once the this or
a little more difficult and

in particular we still don't know how
to differentiate just by looking at

sequences whether an enzyme is going
to be an end for methyl C N sign or

an end six methyl a enzyme So
I want now to talk to you about how

we go about analyzing genomes in order
to find new restriction systems.

So the any genes a really easy to
find because they have good sequence

multi-faith within them and you can think
about this in terms of the fact that

they do two things these enzymes
first of all recognize D.N.A.

but they recognize a lot of
different D.N.A. sequences and

so they will have regions within them that
are very variable because they have to do

with sequence recognition but they also
have a section of the gene protein

that is necessary to do the chemistry and
these are concerned this is the same.

And so
that is makes them pretty easy to find.

Now these as some of the genes
the acid the specificity subunits of

the type one enzyme and the V. genes
are the genes that do the mismatch repair

these are pretty easy to find and so
when you see one of these you can think.

Well yes we must be part
of a restriction system.

Do you see genes the genes that control
expression of restriction enzymes.

Some are quite easy because
there's one big family of

these things the we've located.

But then some are quite difficult.

There are a few that look as though
they've evolved in some separate way.

And finally the things we're
really interested in the our genes

the restriction enzymes genes are very
difficult to find and the reason

that they're very difficult to find so
I skipped that slide the reason that the.

To find is that they're
evolving very rapidly and

it turns out that even when you have two
restriction enzymes genes that recognise

exactly the same sequence very often
they show very limited sequence

similarity in some cases to a point where
you know you would never guess that they

were actually recognising the same
sequence and this is because that part of

host parasite interaction constantly
being challenged by you for

ages and also they don't have any direct

selection pressure on them except
when there is a challenge by afraid.

And if you happen to be in a situation
where you're in an organism with many

different restrictions systems and I'll
show you an example of rat in a moment

provided one of the restrictions
systems is active and

can dispose of the phage coming in.

Then there is no pressure on the other
systems to maintain their activity.

And this is reflected by the fact that
they're changing extremely rapidly.

Now if we go through and

look at the sequence microbial genomes as
of yesterday there were twelve hundred and

eighty five bacterial genomes for which we
have complete sequences and I mean closed.

Sequences not just shotgun data we
have ninety three are calle sequences.

And among all of the eleven
hundred fifty five of them have at

least one restriction system in them.

The ones that don't tend to be
bacterial endosymbionts things that

live inside host cells
many examples of this.

They apparently don't face problems for
from Fader's or

if they do they find out of
ways of dealing with them.

Now a lot of what we do is to go.

Sorry I went the wrong way that
a lot of what we do is to look at

the newly sequenced genomes and see what
we can find by way of restriction says.

Thems and a number of years ago
Helicobacter sorry Helicobacter

pylori was sequenced and it was discovered
that in this organism there are no

less than twenty six different
restriction systems present and

for some reason this organism is just
taking up restriction systems and

what I show here is not all twenty six but
just type.

Two because these are ones
in which we actually

cloned every single example
of these systems and

tracked Frank tippity and what we
found was that among these systems.

There were six that had active
prescription enzymes associated with them

but were eleven that had active
methylated genes associated with them and

the rest of the ones that are active
are the ones that are labeled

H P Y A five H P Y A three and so on.

They have a peer after them.

If that putative that is if they were
just identified by Bioinformatics.

But now a lot of Helicobacter pylori
strange different strains that have been

sequenced.

And so far no one has found
a strain in which the two strains

have exactly the same complement of
restriction enzymes being expressed and

methane transferees is being expressed.

And so it may be that in some way
these provide an epi genetic mark

in the case of bacteria to somehow
prevent somehow identify the species and

prevent cross cross breeding of any sort.

So this is rather an extreme example it's
not the most extreme example but there

are a lot of organisms that would have
four five six restriction systems in them.

So this gets to be quite interesting.

It also makes life a little more
complicated and it also explains.

Why these enzymes are evolving so
rapidly because in a case like this you

know if this enzyme is sufficient
to actually blog entry by the face.

Then there's no selective
pressure on the others and

it's only when a freight comes in.

That is modified against all one
of the systems that you then

have to worry about that
system being active.

It also turns out that a lot of these
systems that have just active methylation

genes have right next to them
a restriction enzymes gene but

is one mutation away from being active and
in two cases we've shown that

directly that you can reactivate
by a single mutation.

OK So there's a slide that was missing.

I just put it in the wrong order.

So I've already talked
about that we can press on.

So now what I want to do is
to show you the results of

an idea that I had while I was
taking a shower one day one of

the things that I think is kind
of fun about bioinformatics

is that there are all sorts of ways in
which bioinformatics can do things that

are a lot more interesting just
looking at sequence comparisons and

it occurred to me one day as I said
I was taking a shower at the time but

when you do a shutdown
sequence experiments so

you know I clone a bacterial genome I
do shotgun sequencing of everything.

Typically you clone two to three killer
bass fragments you take five hundred

base pairs reeds from each A And so
the question is what happens to fragments.

But contain intact non-clinical genes
such as restriction enzymes genes.

So you know let's say one of these two
K.B. pieces had a restriction enzymes gene

and it's well when that goes into a colon
and expresses it will kill the colon.

And so any such genes should be
missing from the set of clones.

But to find the shotgun.

That's illustrated on the next slide where
you can see that if you've got a clone

that runs up here just goes a little way
into the restriction enzymes gene then it

should be perfectly OK there's no reason
to think it's going to have a problem but

as soon as you start to take
clones that start over here and

run into the restriction enzymes gene and
contain the complete gene they're

going to be dead and they're going to be
missing from the shotgun sequence set.

And eventually you'll get to a point where
perhaps you can clone something from

the middle of the restriction enzymes gene
and now it will proceed perfectly well and

of course all the ones
down here will be OK too.

And so we thought well I thought what we
could do is we could look at all the shock

and sequence data and see where
there were gaps in the coverage and

in principle those gaps should
correspond to lethal genes and

in particular in this case
the restriction enzymes genes and

they should tell us when
restriction systems were active.

Unfortunately there was plenty of data for
this marvelous

influenza was the very first genome ever
sequenced it was done by Craig Venter

they had the original data they had to go
down in the basement and gather it got me

the original data and we started looking
and what I've done here is just to plot.

The position of the five crime end of
the READ going from left to right.

And these are the two genes down here
here's the restriction enzymes Gene Here's

the methylated gene and

what you can see is that upstream
of the restriction enzymes gene.

There is a big gap because any clones
starting in that area would have

gone straight through the restriction
enzymes she would have had it intact and

would be dead and if we do
the same thing on the other strand

we see here that there is also
a gap upstream of the restrict.

An enzyme Gene except for
this one guy here.

And when we looked indeed what
was in that particular read it

turned out it was a crime.

Eric.
Clone that had a little bit of sequence

from here but the rest of
the sequence was from somewhere else.

Elsewhere in the genome.

It was just that when they
did the original shotgun.

Two pieces got fused together.

So we looked at a large number of
genes a large number of genomes where

the data was available and where we
knew what was active and what wasn't.

And it turned out this was a very
nice way of finding active genes.

So the next thing we did was to go
to a strange way we never looked

before where no one had ever done the
biochemistry and we looked at the math and

a carcass no metal a carcass in
this case a metal Akaka strain and

bioinformatics we could see there
were three nice genes up here there's

a methylated gene this at the time
was just an open reading frame that

didn't look like anything else in German
bank and then down here was one of these

very genes the thing that is responsible
for mismatch repair and when we look at

the reeds you can see down here there
is a nice gap going from left to right.

Upstream of the restriction
enzymes gene and

on the other strand there is a nice gap.

Upstream of the restriction
enzymes gene in the other way so

we were pretty certain this was going to
be an active restriction enzymes gene.

So we clone did that and we tested it.

This was just we actually tested it
in vitro transcription translation we

just amplified a little bit of D.N.A.
that contained the gene put it

into an in vitro transcription translation
system added some land to D.N.A.

and ran it out on a gel and you can see
here you get a nice banding pattern.

This one is actually
the results of taking just

this in vitro transcription
system from the caucus gene over

here we recognize this pattern actually
And so over here is the pattern you.

Get from an enzyme called B S S H
to a known restriction enzymes.

And in the middle is what happens if you
mix the two just to show that these two

banding patterns really are the same.

And if they were different.

You would get extra bands in the middle.

When we looked to the side of cleavage
we discovered that whereas B.

S.S.H. two left for prime four base

three prime overhang this one left
the two base three prime over.

Sorry the five prime overhang
to base three prime overhang.

So the two ends are to have no sequence
similarity they don't look anything like

one another and
in fact we found a number of

other you restriction enzymes
doing exactly the same technique.

So I thought this was kind of
a funny way of using bioinformatics

to to do things that were not
immediately obvious and yet

actually you get functional
information out of this kind of data.

Now unfortunately everybody's
discovered always hard throughput

sequencing in cloning any more.

And so the amount of data available for
this is gong.

So we can't keep doing this for
a while it was pretty good.

So now what I want to do is to
talk to you about a new project

is just getting started and
it's called com breaks it's

called computational bridges
to experimentation and

it sort of builds on something that has
been a focus in my lab for a long time and

that is I like to do bioinformatics but
I like to sort of make predictions and

then test them I like to link
the bioinformatics to the biology and

I have an experimental lab in addition
to several points from the Titians

who work with me back in two thousand and

four I wrote a paper a little commentary
that was published in past biology.

And it was basically saying that we have
a problem at the time there was a lot of

sequencing taking place we would
gathering D.N.A. sequences galore.

And what we were finding
was that there were high.

Hundreds and hundreds of genes that you
could identify using Mark's programs and

other people's programs you could clearly
see here is a nice coding region.

We have absolutely no idea
what those coding regions do.

And so my proposal because there are no
high throughput methods that we know

of to do this my proposal was that
we do a pseudo throughput method and

we achieve high throughput
by paralyzation and

the idea is you get all the bioinformatics
that you can who want to collaborate on

this to make predictions and put them into
a big database and so we have a great big

database of predictions and then we
recruit biochemists to come along and

test some of those predictions elected
ones things that are going to give you

a lot of value in terms of later
being able to do bioinformatics or

later being able to understand
the biology of the organism and

in this way we would
build a set of genes for

which we had well characterized
experimental functions for

instance it's already known there
may be among the genomes that we now

at the moment there are plenty of examples
of genes that occur in five hundred

six hundred of these organisms and
it's clearly it's the same gene.

We have no idea what they do.

Absolutely.

Unknown function and so this is a way of
trying to find out what the function is to

expand our knowledge so that perhaps
ultimately one day we really can do

systems biology because we know what
all the gene products are doing.

So I say this was proposed
in two thousand and

four I got a very positive response from
an I age they said let's organize a little

conference talk about this and
see how we can make progress go forward.

So I did that in late two thousand and

four I organized a conference down in
Washington all of the key people from and

I came to this conference of all people
associated with the funding agencies.

But the problem was that
what I wanted to do was.

To match up expert
biochemists with particular

predictions that lay right in their
area of expertise and in such a lab.

It doesn't actually take very much extra
money to support a student to come in.

It could be a rotation students could be
a graduate student could even be a high

school student in many cases but or
even an undergraduate to come in and

just sort of you know make a little bit of
the dream product put it through the ass

eyes that are well known in the lab and
find out whether this prediction is

correct or
not because at the moment what happens.

Loads of people are making predictions.

No one is testing them part of this you
know biochemist don't like to test other

people's predictions particularly.

So the idea was well what we'll do is
we'll give a small grant to the lab.

We'll keep them five ten thousand dollars
just in order to test this particular

prediction.

So that was the essence of this proposal
Well nothing happened until two thousand

and eight late in two thousand and eight
when and I each in their wisdom decided

I should as a result of this all appear
always funded said well you know you're

not asking for enough money we don't
know how to give away five thousand but

we know how to give away two
hundred fifty thousand dollars but

we don't know how to manage your
portfolio as small as this.

And so nothing happened.

I couldn't get them to do anything so
in two thousand and eight.

Finally an R.F.P. request for proposals
came at tackling exactly this problem.

How are we going to do annotation
better because even and

I finally realized in
light of the numbers.

You will see here this is the growth in
the number of microbial genomes coming

at this is two thousand and four and
here when I originally wrote my proposal.

They're getting worried they're
spending all this money on sequencing.

But the functional annotation of
those sequences is lagging way

way behind and you could at this
point I think make a strong.

A swell was a point in sequencing anymore
genomes if you're not going to try and

work out what the D.N.A. is doing.

Why bother gathering more D.N.A. sequence.

Well you know obviously D.N.A. sequences
don't find that a very popular idea.

And so they felt I had to do something
about this functional annotation problem.

So I put out this request for

proposals we were applied to it a group
when I say we aren't talking about two

Boston University where I have an adjunct
position and I and wrote a proposal that

essentially flashed at this original idea
that I've had in the plus biology article.

I just want to illustrate the problem and
it really is significant.

So mufflers influenza was
the very first genome

ever sequence this was done back
in one thousand nine hundred five.

And if we look at the content
of the genes you discover that

of the sixteen hundred fifty seven
genes that have been identified.

There are twelve hundred twenty eight that
are either now because someone's actually

done some experiment on them or
you can make a good prediction.

Now what do I mean by a good prediction
multiple clear I mean a prediction that is

based upon some computer looking and

saying here's a gene is ninety
percent similar to this gene and

another organism which has been
annotated in some particular way

the hydrogenated it's an oxidizer whatever
it is and the substrate is known.

However there is a problem.

So what happens very often with
computational predictions is

that you've got a gene that may or may
not be what I call a gold standard Gene

as you will see in a moment where some
point someone actually did an experiment.

And you find another gene that
is ninety percent similar to it.

So you say this must be the same and so

it gets annotated in German bankers
being the same and at some point.

Things go wrong.

This gene that was originally
ninety percent similar

to the gold standard gene one
where the function was done.

Somebody finds a gene that is
ninety percent similar to that and

say it must have the same function too.

And so what happens is the quality of
annotation tends to go down because

almost all of the annotation is being
done by computer programs not by people

computer programs are only as smart as
you tell and they should be very often

things that ninety eight percent identical
can actually have different substrates.

Many examples of glucose or
hydrolysis things that cut up sugars.

But you missed a couple but
you know acids difference and

all of a sudden the sugar that it
will cut next to it has changed.

Instead of recognizing right.

Recognizes silos and
so on and most computer

programs that are just doing all of this
annotation these days don't recognize that

you know ninety eight percent similar
wonderful it must be the same.

And so there is a lot of wrong and
attention in G.M.

bank some people best are made and
there might be as much as fifteen

percent of all of the annotation
in Jenin bank is wrong.

You cannot be trusted.

The bottom line is you
really cannot trust it.

This point since we don't know
which fifteen percent that is.

So evil these twelve hundred twenty eight
genes where there are good put Asians.

There's a fair chance two or
three hundred of these are wrong.

Also the we don't really
know what they do.

Now what's left is four
hundred twenty nine are known

these are genes that we
have no idea what they do.

We're told they're labelled as
being hypothetical protein.

There is some computer program of one
of Mark's programs came along and

said this is the start of the gene and
here is the end of the gene and

you look and sure enough makes
a perfectly good proteins for.

Fifty amino acids long looks
like a genuine protein.

We have no idea what it does it could
be similar to something else in twenty

other genomes probably don't know what
it does in any of those genomes either.

And so that is a bit of a problem and

then there are others that
are sort of hypothetical concern.

So that's what this
twenty six percent is and

this number is actually very close to what
it was back in one thousand nine hundred

five when the genome was first sequence.

There's not been a lot of progress
in understanding what the genes

are coding for.

In some of these other organisms
the numbers are higher

than a carcass was the first our kale
genome done thirty five percent.

No idea what the genes do.

One of the more recent ones
is Pseudomonas aeruginosa

where we have forty two
percent of the genes.

Almost half of the whole organisms genome.

We do not know what it is no idea
what it is doing and yet for

many of these genomes people are trying
to do what I call systems biology.

They're trying to work hand computer
programs that will tell you

how these genes are working
how the organism is working.

I think it's not doable.

The moment that is just simply not doable
but there's no reason to think that it

couldn't be doable and I think we should
focus on at least a few of these organisms

working out exactly what all the genes
are doing and that means doing two things.

It means.
First of all working at biochemically

that the genes are capable of and
then ideally genetically

putting it in within the context of
the working of the whole organism.

But the thing that is easy
to do is the biochemistry

the genetics Often all it ever
does is to give you a hint as to

what's going on maybe give you
a prediction as to what this gene is doing

it's not often that the genetics
absolutely nails the function

biochemistry doesn't a over function
it nails the in vitro function doesn't.

Tell you what pathways it may
be participating in and for

that you ultimately will
need to do with genetics.

But very often just knowing the
biochemical function can take you an awful

long way towards understanding
its role within the cell.

Sorry wrong way again.

So here's the cause of the problem you
know you hear a lot about how great it is

to be able to sequence D.N.A.
for less and less and less.

This is the cost per base pair.

The number of base pairs you get per
dollar and you can see what's happening

that the number of base pairs you
get per dollar is just going up and

up and up is soon going to
be right up to the sky.

It's amazing how cheap it is now
to get D.N.A. sequence and so

of course you know it's simple to do so
everybody does it but I ask myself what

is the point I ask you what is the point
in gathering all of this sequence.

If we're not going to try and work out
what it's doing if we're not trying to

try to understand the meaning
the biological meaning less

so why has there been so little progress.

Well there's a bunch of interesting
reasons one is the inherent difficulty

of the problem if I give you a protein of
three hundred fifty amino acids long and

I say hey I'll give you
a hundred thousand dollars work.

What it's doing.

You can't do it all you
can do is sort of Asif

everything that you think it might be
doing if you're lucky you'll hit it.

The odds are you're not
going to hit it and

so we definitely need to find ways of
making reasonable predictions about what

it's doing does it look as though it might
be interacting with accidental some of

finding does it look as though it's going
to need A.T.P. does it have motifs in it

suggests that it might be involved in this
and this is where the bio information is

can come in and there are many ways in
addition to just looking for motifs or

just looking for a sequence similarity
that you can make predictions I showed you

one of them in the case of the restriction
enzymes genes where you could act.

We identified genes that were lethal from
data that was carried in some completely

other way and there are many other ways
you can perhaps identify operands and

say Here's a bunch of genes five
genes that clearly all involved

in human bio sentences but there's
one open reading frame here right in

the middle of this opera and
we don't know what it does.

Can we perhaps make a good guess
as to what it might doing.

We could guess it's involved in here
by a sentence or two in some way.

And so
maybe there's a biochemical step involved.

For which we don't have a gene
you can go and test it.

So there are many ways in
which you can do this but

it typically takes a combination of
biochemistry maybe some genetics.

Maybe some computation to begin to
get a handle on this problem and

not a lot of people
have all of the skills.

So this cross disciplinary set of skills.

That's needed to really make progress
does not reside in very many labs

a lot of labs just focus on computation
others just focus on biochemistry and

a lot of them never talk to one another
and in fact a lot of chemist don't want to

talk to the bio information because they
think they're very upper to make it way

too many many predictions the other thing
that's happened recently is there's

this tremendous appeal for genome wide
studies everyone wants to do something

where you look at the whole genome all at
once you know let's not worry about these

individual things let's look at everything
and in fact it's becoming increasingly

difficult to get grants in areas where
genome wide studies are popular.

If all you want to do is actually look at
a few genes study sections say why don't

you look at the whole genome Why don't you
do micro rays Why don't you do this other.

Well unfortunately there are no genome
wide ways other than computationally

of getting this process this
problem to function and

then finally there's a lack of appropriate
funding mechanisms you cannot go to and

I can say hey I found an interesting gene
here it's present in five hundred genomes.

I want.
Work out what it does no one will give you

money for that you simply can't
get the money to do that.

So this is where I come bricks comes in

through put by paralyzation so
we get the bioinformatics sions on board

to make high quality predictions
things that really say.

I think this is an oxidizer and I think
the substrate is likely to be a sugar or

I think the substrate is likely to be
a turkey and I think it's likely to be

an alley frats it can mean an hourly
phatic substance of some description so

basically we want to get some high quality
predictions we want to assemble a database

of those predictions and we're in
the process of doing that at the moment.

Biochemists can then come and
test the predictions and initially we will

approach biochemists when we have a good
set of predictions where on the Web site

assembling a list of what we think are the
top one hundred per predictions things

that if these predictions turn out
to be accurate and can be tested and

shown to be accurate then that will
be very helpful to the information

in making better predictions later on and
so that is something the an important

criteria here because you know you write
an algorithm you make predictions.

If old predictions are correct you
think are the algorithm is great but

if half of them are wrong then you've got
to go back and readjust your algorithm so

that you can get the right answer.

More often.

And the idea is that by
getting biochemists and

having them work in their field of
expertise the incremental cost of doing

this work is really very very slight.

And so even though it's quite sly you
still have to persuade them to participate

so we offer them some financial
incentives to do this

given five ten thousand dollars and
the way one can think about this going.

Is that maybe there's a student
here at Georgia Tech maybe one

of you in the audience who says.

I wouldn't mind doing this as a rotation.

During my first year as a graduate
student and so I will go and

look in COM bricks and I'll see where is
an interesting gene something we don't

know the function for there's a very
good prediction you look you find it you

say all is a professor at Georgia Tech
who is actually working in this area.

And so between as we can apply to
combat for a grant perhaps get five or

ten thousand dollars so that I can go and
test the prediction in this person's lab.

So this is very good for
the lab to get a little extra money.

Very good for the student because if in
fact they can validate that prediction.

You can publish that result will be
a publishable result and in the process

you will help both the bioinformatics and
the biochemical communities so

you do a very valuable function
a very valuable service for

biology as a whole and the nice thing
about this is that there are so many of

these genes I mean there are thousands
upon thousands of these genes but

no one has to do exactly the same training
everyone can do something different.

You can all make a contribution and
it can be really very valuable and so

we're hoping to get a lot of momentum
within the community to do this

a lot a lot of opportunities
a lot of possibilities.

So one question that comes up had can
we improve the predictions Well I think

one of the things is we should recognize
the vast bulk of the current annotation

is really based on sequence similarity
to previously annotated genes.

This is OK but there are some problems
with it that I will point out in a moment.

We have to recognize a lot of
the annotation is wrong for

exactly this reason I quoted earlier that

something is ninety percent some similar
something we know the function of and

then something is ninety percent
similar to that and ninety eight.

Scent similar to that before you know it
you're ten percent similar to the original

gene and
the chances of it being correct a zero so.

We have to do provide a much better
basis for similarity based annotation.

That will be the subject of
the next slide and then we

will provide within the complex web site
a means for reporting Messina tensions.

What.
All of this boils down to is that

we need a gold standard set of genes and
proteins we need to

have this set where we can say we
know the sequence of this protein.

We know the function of this protein
because the biochemistry was carried

out on this protein with this sequence and
then have a standard we have

a gold standard against which we
compare every new gene that comes into

the database we can say how far away
is it from the gold standard and

at some point it's sufficiently far away
you say well maybe we better just recheck

make sure that it still has the same
function it's eighty percent

similar maybe we should check maybe
it's not really got the same function.

So the question is Where would
you go with such a gold standard

set where you might think you know
we've got big data massive databases

we've got gem bang we've got the D.N.A.
database in Japan.

We've got you know prop which is
supposedly the database for proteins.

We have the whole I Database Gene
both of which have huge

lists call I.D.'s and functions
associated with them and you might think.

But these would already be gold
standard databases they're not.

If you go to any of
the right databases and

you can we find out for
this gene where this function

was determined was in fact
the gene sequence upon which

this function was determined we discover
you can't even find that information.

This is not something that needs databases
of considered worth tracking and

what's happened in the.

Color is that perhaps someone in
some in color strange determined

are this is the D.N.A.
methylation D.N.A. methylation E. coli.

But perhaps the function
was actually tested

in a strain that was a long way away
from anything that's been sequenced and

in fact the gene was never
sequence to the time.

And so on and you discovered a large
number of the whole of proteins

fall into this category the sequence
that is in the databases this

ecology in may not be the one where
the function was determined and

so and the same is true you know crowd and
in fact

I can tell you all of these databases
are rather embarrassed by this fact

because when I asked them this question
everybody said no we can't provide that.

And when I proposed this idea for

a gold standard database
everybody said What a great idea.

We should get this.

And so now they're all busy
cooperating with me and

with Combray some with us to
actually assemble this data set.

So this is the essence of it but
each protein in the set.

We're going to find a reference that
describes the determination of function

going to make sure that the strain
is well defined and noted.

And we're going to make sure especially
that the protein sequence is known for

this protein in which the function
was determined if it doesn't for

if we can find this match
between a protein sequence.

It will not go into the gold standard set.

So we're putting this together it will
probably take a couple of months to

get a reasonable start on this because
both N.C.B.I. And you know products

are busy working out procedures to make
this easy but we have got started on this.

We have about a couple of
thousand candidate genes and for

each of these genes that being sent out
to manual curators who would go along.

Through literature and

I actually find whether or not
the literature supports the assignment.

So we identify candidate side from
databases or from individuals and

we welcome anybody who wants
to submit a gold standard

what they think should be a gold
standard on the complex web site.

You know prot will then make sure that
there's a unit product number for

this they will go through their annotation
and prepare a template that in principle

contains reference to the sequence
reference to the function and a few other

things about the strain designation and
so on and then these are being sent

to curators this is something
we're doing as part of the complex

project at the moment but ultimately
will be taken over by and C.B.I..

And what we're asking is just the curators
manually check the strain information

is accurate and sometimes people say well
you know we did this on the color line.

Well that becomes unknown strain.

It's OK that's fine but
one should not therefore assume that it's

the same as something that is
in the sequence databases.

We want to make sure that the biochemical
characterization is accurate

description that is in unit really
matches what was done and then

the fact that the gene was sequenced and
hence the protein sequence is known and

from exactly the strain in which the
biochemical characterization was done and

I just put a note here you've
got to know you know the sort of

the gold standard sequence we call lions
for a strain M G six hundred fifty five.

There has not been very much
characterization done on genes

from the strain so
it's always important to know where it was

to make sure the sequence is
the star the same as M G.

Sixteen fifty five before the annotation
is put in as gold standard.

So how can you help.

Well there are many ways in which
people can help you can send

in candidate gold standard genes and

proteins you can volunteer to serve
as a manual curator you just.

Rich Roberts Roberts said anybody dot com
and you can browse the complex web sites

and let us know if you see genes should
be gold standard the complex website

is complex you got edu again if you
google complex you will find this.

Now how are you going to prioritize
Gene since I told you you know

there are hundreds of
thousands of these things.

Well the first thing we want to know how
many organisms contain this gene for

which there is a prediction and
the larger the number of genes and

especially if it transcends
kingdoms the implants if it's in

animals then that gets to
be a lot more interesting.

So if any gene that is present a large
number of organisms Obviously if you

can work out what it does you
get a lot of bang for the buck.

We're also very interested in trying to
finish the functional annotation of E.

coli and so if the gene is
currently unknown in color and

if there is a good prediction and
that would be a high priority for testing.

Something for Helicobacter pylori we
wanted to show that this approach could be

used to annotate pathogenic
genomes obviously enter and

I aged very interested in pathogens.

Aleko back to Laura a very interesting.

This is the organism
the course has also but

it's also one of the few bacteria
absolutely known to cause cancer.

It causes stomach cancer.

It also causes this off.

Adriel cancer.

It can be grown in the lab and it's
actually relatively easy to work with.

Is there a clone or purified protein
available the structure of you

know makes initiative has actually
generated many thousands clones

of interesting unknown open reading
frames that are present in G.M. by and

the idea there was that they would say
well let's make this protein Let's try and

crystallize that if we can get
a structure then perhaps we

can determine the function well it turns.

When you try that approach about
one in ten of all the proteins

you make actually crystallize and
then when you get the structure.

Most of the time you haven't got a clue
about function structure does not easily

give rise to function but all of these
clones and all of these proteins were

viable and we can provide those we
can tell you how to get those and

then there's a whole bunch of other
criteria I don't want to go into so

there's a whole list of things that make
it interesting for particular proteins.

Who can help.

Well the computational biologists and

the biochemist obviously
also geneticists can help.

Finally university students high
school students who are actually going

to run an experiment with our local high
school to have their students come in

to buy a labs and do some of these tests
as part of a high school fair project.

I was at the Intel Science Fair earlier
this year and I can tell you the kids who

were participating there from all over the
world are more than qualified to do these

kinds of tests if they go to a decent
lab and get some supervision and

I think you know you could easily go and
find high school students here

who could come and work in labs at Georgia
Tech that would be straightforward.

You know here professors there
are a lot of professors who retire and

go emeritus and
they love doing experiments.

Many of them take the opportunity
to go back to the lab.

Not looking at anybody in particular but I
think there are some opportunities here to

find finally of course we need
help from the funding agencies and

I've been talking not just
an eye to N.S.F. to him.

I have a Hughes for
the Welcome Trust in England a lot of

them of expressed interest because this
can actually be a wonderful teaching tool.

There are many ways in which you could
build this into undergraduate education

and many ways in which you can
run some of courses in which you

get students to participate
in something like this and

a lot of times you don't need
a lot of funding you know.

To get it going.

We're hoping to make this
into a worldwide effort.

It's an ideal project for developing
countries where they don't have a lot of

money for science but often there is a lab
that has a lot of expertise in an area and

can do this and
can use this to train students.

So I think there are lot of possibilities
for people to get involved in this.

And of course you can help.

These are the people that
are involved myself Simon to see if

who is a computational biochemist
than a Boston University

a guy who came out of computer science and
moved into Bioinformatics.

Martin Stephan is a bio chemistry and

he's looking after a lot of
the biochemical side of things.

Charles de Lisi is
an emeritus professor be you.

He was the guy who first
provided funds for

the human genome project
when it got started.

Daniel said Gray is a young assistant
professor at B.U. who looks at networks

he's interested in networks with
in bacteria and other organisms.

Stephen Saltzburg is a well known
computational biologist at the University

of Maryland.

Denis picked up again another well
known Byron from attention at Columbia.

All of our collaborators we already
have a lot of collaborators both on

the computational front and
on the biochemical front.

We have half a dozen projects already
funded to get going on this and

of course we're very grateful for
funding from the and

I ate they gave us a to get Grant on the
stimulus money so this money is going to

run ads and we're busy working out how to
get them to give us a lot more because

obviously we're just at the start of this
but I think the very exciting project.

It's something in which we can show that
we as biologists really can collaborate

because however you look at it.

Science is a collaborative
activity you know everything we do

builds upon stuff that's
been done elsewhere.

And I can tell you it is a lot more fun
to collaborate than it is to complete

a long standing collaboration with
my colleague here at the front.

Really and crystallographer we've worked
together for many many years and it's so

much more fun to collaborate than it
is to try to compete with people and

I hope that all of you will but
at least think about this and

see if there's some way in which
you can help and collaborate.

Thank you thank

you very much for
the inspiration to get the ball.

I'm going to buy over the place
in sequence is the book

to me which is increasing

all your problems and

you should see some problems.

This is your question my

letters for you with my

eyes help you need to construct to convert
double standards or anything else.

Even your heart.

I read well.

So I think first off let's
say you have a protein.

If you have a gold standard.

There are many ways in which you
can say how far away it is so

obviously just a simple
statistical test and

many amino acids are identical but
also you can look at motifs are key

motifs conserved in particular if you
know that within the gold standard.

There is a little area that forms
a catalytic sites all such substrate.

Condition site and

you can see that that is different
in this new gene there already.

You can begin to say well you know maybe
this new protein doesn't have the same

function maybe it's different.

Many of the programs that are annotating
genomes at the moment don't even

do something.

The simplest and so I think the idea
of providing the gold standard is that

it will really give the annotator.

The opportunity to do a much better job
of annotating things because they have

something to relate it all to and that's
not what is happening at the moment.

You know there are more of one

on one.

So we don't want to call it works

you call a cute little guy so

expression is actually a lot
easier than it used to be but

there are a lot of tricks and
I think one of the key things is you

really have to keep your eye out for
where people

have come up with good traits now one of
the things we're focusing on bacterial and

genes not doing any eukaryotic genes at
the moment in this particular project.

Partly because of expression but partly
because you carry out it genes are so damn

difficult and complicated even predicting
where they are can be a big problem.

I mean I'm sorry I discovered this split
gene thing because it just complicated

biology enormously.

But I figure if we can really get
everything going with bacteria or

an R. K. or and get a good group of
people together and get everything going.

We can at least show that it will work and
then we'll expand it into you carry out

later on so
that is the idea at the moment.

So call out in which we know that
you know it's nice that in message.

You know.

So we're not looking for
cellular function.

All we're trying to do in this particular
project is to get at the biochemical

function because if you want to go beyond
that you're talking about really a huge

amount more work and not something that
can be funded for a spall amount of money.

What we're hoping though is that what
will come out of this is that people will

find some new functions that can then lead
them to right are ones to explore it and

find out about the biology what that OK.

And what's right.

So for protein protein interactions I
think we would be satisfied at this point

if we could identify the binding partner.

So for
instance if it's a transcription factor.

What is the D.N.A.

sequence that it binds to if it's
part of a multi subunit complex.

What what is it a part of the right to
some of proteins you know that part of

a massive complex some of them
have enzymatic function but

a lot of them don't.

But just the fact that they're a part
of that complex is good enough for

our purposes at this stage.

So we're not trying to solve the whole
of biology we're trying to do something

pretty simple that will actually lead
to better annotation of genomes so

that's really what it's all about you

know

right.

So I told you we have this class of
enzymes that are recognized by represented

by M M E one this is a single train
restriction modification enzyme and

in this particular case we already
know how to do that we can make

many changes specifically in that
particular class of enzymes.

We have another class of enzymes that
we're just starting to make some progress

on where we think we will
be able to do the same.

Sort of thing and

that is our go buy a lab says not to just
keep offering more and more enzymes.

But to say you want an enzyme
with a particular recognitions

sequence specificity Telos and
we will make it so this is our goal but

most of the work up by a labs now
is our aim towards understanding

enough about specificity so
we can make new ones by design.

Yeah.

You don't.

OK Just a scratch I'm
sorry this is working for

first class restriction enzymes
structures from you know for sure.

Structures and watch.

It's a very mature for the woman so

that fall into about four
classes you can classify them as

four among the ones we know we don't have
a lot of structures we have right things.

Sixteen or eighteen structures at
the moment and we certainly are certainly

not fully representative of the range of
restriction enzymes that we know about and

in particular the thing that I would
like to see some structures for

examples of enzymes that recognise
exactly the same sequence and

cut it exactly the same point.

And yet show no sequence similarity and
we in a while you

have one such example of that although
it's not quite that Eco are one and

bam H one recognise different sequences
but the fall that almost super

imposable the structures and the only
three amino acids that they have in common

are the three that form the catalytic
triad that do the cleavage and

that's not of any use in terms of
predicting stuff which you are clearly

one among young people five

years we've got to see because they're
pretty much reacting in a fine.

Gesture to a late sense.

Right right.

And we have tried some of the.

At some cases you can make

predictions about what these
things are going to recognize.

But in general that is not that approach
has not worked terribly well with

the sequence differences
are too different.

So for instance I go I want Bama
each one which almost coincide.

If you try to thread one on
the other it doesn't work

the threading algorithms
don't get it the same.

So it's a problem structure protection in
general is a bit of a problem if things

are close and threading works very well
but as you get further and further away.

I mean lots of these have sort of you
know four percent similarity of props and

identity and that it's quite a problem.

There's that we be glad to send you

some if you want to try reading signals
the functional size trying to use them.

Why this mean.

Terrible shock if you want
to actually write but

the problem is that the catalytic triad in
Bam makes one an echo of one you've got

one here one here one here in
the sequence in space right space.

Right.

See that's the way it goes and so

you get at least in theory which might be

OK Well we'll send you some to try and
there was you one question I think.

Yes OK he's

right.

If

I was running and
I would do that but I'm not OK.

Yes.

Membrane proteins are a major.

Well I'm from every conceivable
function that you can imagine.

So from a structural perspective
from a functional perspective and

I think at the moment we're going to leave
the membrane proteins to the experts So

for instance some of them.

We know are transporters some are involved
in transporting things to and fro and

we really Milton Sawyer is
a collaborator of Oz and

so he is going to deal with that
aspect of it in some other cases

we know that they're just
integral membrane proteins.

You know what they may be interacting with
what their next to what kinds of things

they bind and so we would just do
the best we can but they're likely to

be the very last proteins that we have
a kind of sign a decent function to but

even knowing that they're
an integral part of the membrane and

particularly if you have some idea
of which part of the membrane they

are involved in that can be helpful and
you know.

Biology is incredibly complicated.

I mean you know these people who think
they're going to understand the nervous

system I say you go back and
figure out how some simple bacterium works

bacteria a lovely day a beautiful
beautiful organisms and

we should be able to
understand how they work or

by that I mean we should be able to write
a computer program that will simulate one.

But we have to know an awful lot more
about them before we can do that and

I will guarantee that once we
understand have bacteria work.

We will have a much better
idea of how humans and

other you carry out its work but
we believe are complicated.

It's quite amazing and it's great you
know because it means we've got jobs for

years to come right thank you very much.

and
so he is going to deal with that
aspect of it in some other cases

we know that they're just
integral membrane proteins.

You know what they may be interacting with
what their next to what kinds of things

they bind and so we would just do
the best we can but they're likely to

be the very last proteins that we have
a kind of sign a decent function to but

even knowing that they're
an integral part of the membrane and

particularly if you have some idea
of which part of the membrane they

are involved in that can be helpful and
you know.

Biology is incredibly complicated.

I mean you know these people who think
they're going to understand the nervous

system I say you go back and
figure out how some simple bacterium works

bacteria a lovely day a beautiful
beautiful organisms and

we should be able to
understand how they work or

by that I mean we should be able to write
a computer program that will simulate one.

But we have to know an awful lot more
about them before we can do that and

I will guarantee that once we
understand have bacteria work.

We will have a much better
idea of how humans and

other you carry out its work but
we believe are complicated.

It's quite amazing and it's great you
know because it means we've got jobs for

years to come right thank you very much.