Today I'm pleased to say that we have
distinguished guests from Brasil the first

or that she responded you know he's
professor why is the physics of versity.

Legal in us and

she also did need to go.

Obviously this was rubbish.

The rest of us are still
out of the shop students

are in the national the senior
national biostatistics society.

She is on board of directors and

Secretary of the library for
Brazilians obviously got issues.

She was last year.

It's less than the organisation
Committee I've seen up to date which is

the Syrian National was you more
than refuse to use it for THE LEAD.

She got me to the University of
New Kalina job if you are to be

rational about the pace and
in France last year I was with huge

research volves I was the V.C. got you.

Shoes the modern nation
our interview process but

now here is your solution we can say our
life is gone they don't mind must leave us

of genetics and today she will be
talking about how much a native.

Does Mogo spot.

Sickness.

Thank you very much for the invitation.

It's a pleasure to be
near her tried to tell.

Make a small introduction of biology.

Even though I think I
don't need to hear you.

Usually the audience is
more just that delusions.

So I'll go first over it just so you can
have an idea of the type of data set.

I'm working on.

OK I will talk briefly about the him
in distance to the compositions.

And then I will talk about the homogeny
to testimony groups of D.N.A.

simple is when you have a certain
kind of data set up for

each individual who have
a seek with a D.N.A. sequence.

Then I would talk briefly
about the distance measures in

Meeker satellite data which
is another kind of data.

I will briefly explain the two.

Different kind of doing this that we can
work with then whom would you need to test

with this kind of datum because satellite
sequences in our all explain briefly.

Mission to apply cations.

So.

If you have.

The D.N.A. a D.N.A.
sequence you know that you.

You have a double Well X.
sequence is always the T.

pairs with A in C. pairs with D. in if you

are a deployed individual you
get like a pair of sequences.

So you have a C. T. G. Your T..

So the type of data that
I can I will get from

each individual would be
a sequence of letters.

Like this sequence is of letters.

The difference with satellite data is that

in some part of the genome
that it's not caught.

Did for a protein or.

You have been to an annex
owns Of course you know in

the part that it's not called
that sometimes get a bunch of

letters that doesn't count for anything
but it seems that has a lot of variation

in people are starting
to wonder why it gets so

much variation in why if it's
associated with some disease or not so.

A nick or satellite data or
a variable number of ten then repeat

what you have is repeated D.N.A.
composed by tool to six base pairs.

Like this one this one I'll have eleven.

A sees a Seems they see like if you have
to base pairs that are repeating this one.

You have six repeat Asians.

Like C G C C eighty.

And this is repeated six times.

So in need of satellite
data of what I will see for

each individual in in the certain
locals what I would like to see.

Is.

Thank you is how merely

repeat th NS I have in a certain locus for
each individual.

OK.

So first thinking about the D.N.A.

sequences only the letters then briefly
talk about the meter satellite data.

If you have a vector X I I
index in the individual.

So for
each individual have a sequence of letters

can be a C. T. your G.
in this case we have four categories.

So the hem indices would
be when you look at a pair.

Sequences for
individual are individual Jay.

Would like to see what is the difference
between these two individuals.

So looking at the pair of sequences
to him in distances would be

the proportion of sites in
that sequence where X.I.I.

next day you first saw the Mindy's so

the measure we're talking about here is
the distance measured within the group so

within a group will make all
the pair wise comparisons.

If we have a group with N. G.
individuals for and group G.

will make all the comparisons
between the individuals so.

This would be the distance measures
between groups within group I'm sorry.

So far between groups.

We all compare an individual from Group.

J G in and
intervene drop from Group G. prime.

OK And then we'll make all
the pair wise comparisons between

one interview with the individual from one
group an individual from the other group.

And you see what the proportion
of difference you can get.

And divide for all the comparisons you can
make which is indeed times in G. prime.

OK.

So if Ill have to get the ball say a ball

or have they hear

from Senators remember from our own Candy.

So if feel.

Get the pulled sample make all
the comparisons between individuals

disregard in the groups you will
have this what I called again.

The bar in zero.

This is if we make the comparison
disregarding the group between

all the individuals we have so

in we have any individuals in
the total for the pulled sample.

OK So but this we can Deacon polls as a
measure with Dean in the measure between.

So for the homos in it to test
what I'm going to do is to

test where they're
the population the ages for

the him in distress for the population
in the D's are for the sample.

OK so
what I mean what I'm trying to test is if

the distance that I have between groups.

Can be expressed as
the distance with the in

between like the average of this so

far of the not hypothesis
would be that the the if

there is no difference between
groups that this measure

between groups would be just the average
of the within groups distance.

OK in the the alternative

hypothesis would be that this is
greater than this because of some

mathematical we see that it can
only be greater than or equal so

that's why the they are turned
out to hypothesis it's one sided.

So as we can doing some math.

We can write down
the between measure as this.

OK the twice

the distance is with the in between
groups in then by those each with them.

Group.

So doing with some results
of use statistics.

We can get there.

Syntactic distribution of
this the statistics in we

can actually do some
appley cation with it.

Who say later on that

in my appreciation we can even if we
don't have to bother about this and

we see if I can work with
both who commit fraud here.

We don't need to bother with
the distribution which is a linear

combination of guys squares but
we don't need to bother or

you just do re simply in techniques
like Bootstrap Project knife.

To do the test.

And here is that in the other
situation if we have you

stead of have in the sequence
of letters you have and hope for

each individual the the letters but
you all only going

to called the number of reputations you
have in each side for each individual.

OK So why here.

Why G T It's counting the number
of units of reputations.

Of the I have a copy in.

The death group in
a specific locus at time T..

OK So we hear

there is a trend on the run.

So here I'm just looking at
the difference between two copies

copy I in our prime from
the same group in here.

If I look at different groups of K.
was in the same idea of

the him in distance we're using so
he can have now this distance.

The difference here is that.

Looking at MC or satellite data for
deployed individuals.

We're always going to have two copies for
each individuals.

So that's why I have a sample of
tea in groups because if I have

enjoying the individuals in Group G.
I would have two copies for

each individual soiled have to copy So

OK here I am just looking at
the difference squared as a measure of.

So the distance measure is for me.

Chris that the light data.

Now they are similar to
the him in distance one but

here the difference is that in the him
in distance one we are looking at

categories and
see if they are different or not.

Here we since we are measuring
the number of repetitions.

Now we have a point that of response and

we can just look at the difference
since in the square it.

OK but
the same thing goes on here with the.

These for the pulled sample in the can
decode pulls the pull sample distance

in the with the in between distance.

And.

OK.

The same thing happens here with
the we decompose the with the in and

between measure in

we are trying to find out their syntactic
distribution of these statistics.

The goal is to get a statistic with

an esteem started distributing that we
can work with to actually do the test.

So all the homogeneity
testimony satellite simplest.

We can work with this.

Parameter here.

Which is actually the divvy us in
the expected number of mutations due

to the population diverse
because here we have to work.

Also if I don't know if you have
worked with color sister you or

phylogenetic tree just see
how the evolution goals

in in a species or a genius.

So behind all of that theory.

I also take that into account.

That's why I had that T. that counts for
the time because at the time goes

the genes kin evolve toward the species
whatever we are working with.

So.

The the task would be if that if
the the divinest in the expected number

of mutation due to population diverse and
divergence in a specific local say L..

If the.
It is equal to the devious in the number

of mutations within population and
I am under a earned their no hypothesis.

I'm saying that this is zero.

OK so the test that district
would be this very similar

to the one in the human
distance just the way that.

You have the better
are a little different.

I'm going to skip those formulas because
this is just a simple theory and.

With own the here already
with this right now.

So let's go to the to the application
which is the more interesting part.

For the first the plea creation
less think about D.N.A. sequences.

So I would have Case one.

With eighty seven sequences
which we have fifty

eight homosapiens in twenty nine other

primates the goal here is
to see if our test works.

We'll black to know if
doing our test if I can

actually find difference between
these two groups of K. so.

Their interest is to find if
the not all have partners

would be that there is no
difference between the school.

If we reject that hypothesis.

So it means that

it seems that our attests is working
exactly detecting the difference.

So to say that.

We also got another case where we have
fifty eight this fifty eight sequences.

We just regrouped into two so
groups of twenty nine think with each.

And.

In this case we did two
thousand bootstraps.

We simple.

Took two thousand times in case three
is the same S.K. still with just

the most rap is with ten
thousand is a huge issue.

So it might be nice if

not this fifty eight sequences
before I started the thread.

I divide it.

But under no hypothesis then
I have to the distribution

the Puca distribution would be in that
under the not hypothesis on the nor

her part as I don't know which group
they are they are in the board sample.

OK but then I I would do the test
knowing that they are for

one is from one the one set is one
group and the other I separated first.

Then I will do the test.

So these are the observed values
of the test at districts.

For each case.

The numbers doesn't mean that much but
if you.

Now look at the the believe their use.

You'll see that for a case once
when you do two thousand bootstraps

you reject the hypothesis of
home with unity of the groups.

That's what we wanted to detect that
the groups were actually different

it was the homo supposition
the other primates also all.

But if we do ten thousand bootstraps.

It's the same thing.

We also reject in for cases two and

three and we are not supposed to
reject the hypothesis so the P. value.

Is being it means that we're in
we don't reject the hypothesis

of who was unity of the group because
they're actually the same group and

just divided to see if
the test was working so.

This is just the difference in for

case one the imputed distribution for

two thousand and ten thousand there is
not much difference so it's just a C.

in there you don't need that much
bootstrap to in order to your test words.

So that the second application.

Is.

That data from the genetic analysis
workshop I don't know if you have one.

Already about it.

It's the workshop in
the United States that.

Every year I think every
two years they have saw.

For each workshop they
send you a set of data and

you can work with that data and
then go to the workshop

the way you want each person in each group
on the lies that they know according to

the goal that you would think would
be nice to work with the data in.

Later on this data is public and

you can use for any purpose.

You weren't so

I got the data from the this
D A W workshop number fourteen.

In a saw that the dataset would work for
in the methodology we were working on.

So this data is.

From a study about alcohol in they had

all the satellite data according to

ethnicity in alcohol Mincey so

they had the whites black in others and in
the others they had his spending the and

the the others were small groups so
that's why you group them here.

Then they they also separate are separated

by if they were affected if
they were considered alcoholic.

This are these are the unaffected.

There are non alcoholic
with no symptoms in.

These are the pure unaffected.

Because they never drank
the nice these this group here.

The head drink but they were not
affect them they had no symptoms so.

We had a total of two hundred nineteen
McNicholas satellite low side.

Were this data set in the interest to

verify whether the groups defined by
ethnicity and level all of our clues in

our home was genius at each
locals this test here.

Does for each locals.

It's doing a test.

So you have you are testing
at the specific locals.

If there is variability or
not among the groups.

You can do for the home.

There is a way of doing for the whole data

I mean just Eddie now up in taking
the average over almost side.

But since each local recital.

Satellite data has a lot of variability.

They say one can you don't
see the difference in each

locals specifically if you
were Darkman take the average.

So that's why our test is taken for
each local So because of that.

Sorry.

We had to correct the significance
level here because we are doing several

tests for each local So
we do one so it's like doing

multiple comparisons like in the nulls of
various when you do multiple comparison.

So that's why we've corrected Bible for
the significance level here but

more from me.

So here.

Notice that when you.

We want to know the difference in groups.

We have three go three
groups of that nice to hear.

And three groups of our colleagues.

So we don't have two groups anymore like
we had in the first oblique ation so

we notice that jackknifed works better
when you have more than two groups we

have done some study in where
you believe we found out

that when we have more than two groups the
best thing to do the resembling technique.

Better to use is jack knife.

That's why we use jack knife here.

So here are the.

It's a ninety five percent confidence
interval for our test the district.

Looking at one specific law
course is this slow course.

Over here.

So we can see that.

According to the course of this
interval you see the zero it's way off.

So the zero is not in the interval so
we detected some differences

aren't showing here just the ones that are
more interesting because we did a lot of.

Test with many different laws.

So this is you can find that there
is a difference between the group

of the ones that are affected
in the non affected the and

affected with no seen tons and
this is the never drunk with power or

with the affected in

this is the never drunk in the In
effect and with no symptoms.

So at this locus we could see that
it seems that there is a difference.

Between those groups because

after you see if there is a difference
between the among the three wound.

Then you have to make the move to
Comparison Test between one in the other

and that's where we corrected
the significance level.

For this one.

This is another locos

we could see there it is we
know this big difference here.

But this one is below zero.

Well but there is a difference here
as well but not as much as here

or here maybe but
it's something that someone that knows

more about the data has to discuss about
it in see what's happening in this locals.

Also this one is for
differences in ethnicity.

You see that there is different is
a big difference between blacks in

others also blacks and
whites in white in others.

Remember that on the show it's not that
there is difference everywhere I'm showing

just the last side that I
found some difference here.

The variability is much greater
you can see that the confidence

intervals here are wilder.

So in that specific locals
the variation is high.

But you can still see some difference
between blacks and whites but not here.

Blacks and others.

You cannot see the difference anymore.

It is neither here only between blacks and
whites here.

Here and you can see a difference
between blacks in our various but

not really between blacks and whites in
you can see between whites and the others.

Again this is another locos blacks
in white there is no difference.

And we could see that if we
had many different sept of.

This of figures like there but

them showing just the ones that we
could detect more in difference.

According to the locals you
can see the thumb locals

are much more variable than the others.

So that's why we were doing
separate tests for the locals.

So here are some of the references.

This is about coalescent theory.

This is about you statistical theory.

Bork.

This is the paper.

That uses that data set that up
the geisha in with the homo sapiens and

primates and that's when we develop
that that test for D.N.A. sequences.

These are all for a synthetic results
of the test the distinct fusing used

to sex and this is the last one
the one with maker satellite.

This is probably appearing in it was
accepted in two thousand and ten.

So it's already online but
it didn't appear in paper yet

if you can get online already and
I think that.

Thank you very much.

Thanks for all reference.

On you.

Let me ask one question you
said you were at all of that on

that you have this what it
was you just didn't say yes.

How was school by Clyde saying something
about how it should be this so.

So are these various or or or
have you got rid of residuals.

Yes because the overall US has
like five percent confidence

of five percent level of significance if
you do the both are on a test you have to

divide by the number of tests you
are making if you're making like if

we have three groups of
three three groups and

you're going to see if group one into our
different one in three two and three.

You're making six comparisons.

So you have to do.

Vied the level of significance

five percent divided by by six so
to be significant.

It's harder is if you go.

Yeah it's much harder to get significance
when you divide the level signal

does by five percent because instead of
all having a big value of point zero five.

You need a P. value of point zero five
divided by six to be significant.

OK I don't know if you follow just to

try to explain the ratio between what.

Why is there.

Can I turn off one of

the this this.

Or just what's the difference.

When you combine Yes you can.

Yeah there are tests that in the paper
I make a distro a discussion about.

If you want to combine the test
because that was one of the questions

of the referee.

Yes if we want to combine or
the low side of what we do do.

And so on.

So there is a discussion about it.

With

Thank you.