Philip Dei
Tue, Jul-15-03, 06:12
In my continued effort to provide flamebait for this group .
. .
I have taken a rather, in hindsite painful task.
I have created a program that runs monte-carlo similuations.
Currently I have 12 xlinked popstudies that can be compared
with randomly generated sets based on different assumptions.
The initial runs gave pleasing results however a couple of
issues have bothered me.
Since I figured I had 12 individual author bias would be drawn
out by averages. But that the results of a couple of these
Il2rg and Ids loci would never be hit even by monte carlo.
This is problematic since Monte Carlo analysis could be
imployed to look at specific deviations, such as a particular
MRCA that is say in the 99.9% of ages. IOW I might have 10
loci that fit perfectly and have 2 that are so far in left
feild . . . .
All was going well until I ask the basic question. What is it
that we need to get to know exactly when an MRCA was.
1. The mutation rate (exactly)
2. The precise average of substitutions off the seqMRCA in the
current population.
I have just finished one Loci and I have found all kinds of
things that cause worry.
Plp locus 2 variant positions in humans 50% of sample have 1
variant derivative while 50% have seqMRCA. The difference
between the seqMRCA is 5 to chimpanzee, 10 for gorilla.
The binomial distribution on mutation rates based on 5
sequence differences ranges from .031 to 2.24 times the
calculated rate. For 10 it is a little better. The 96%
confidence range for relative rate differences between chimp
and gorilla is from gorilla evolving 5.5 times that off chimp
to 0.45. Thus these rates might apply to an evolutionary
segment like humans despite the calculated mutation rate.
This basic analysis has determined that a low number of
mutations between chimp and humans gets refactored into the
analysis on several levels. The variants at the loci may not
be reponding to the observed rate of mutation but a real rate
that is faster or slower than would be inferred from the
chimp/human absolute differences if those differences are
small. If there are 5 differences between chimp and human we
assume for instance that each difference represents
3.4 million years, and that since each individuals has 0.5
derivatives then the MRCA is 1.2 million years. But the
variance of potential differences of chimp and gorilla rates
could mean that the difference represents a few hundred
thousand or millions of years.
The basic bottom line. In order for MonteCarlo to be able to
compare human Derived MRCA versus the MRCAs expected in an
idealized model the human factor needs to be factored.
This factor can be broken down
4. Variance in sample size
a. Its effect in predicting the depth of mutations at a
site, and thus range of variation that could explain
observed values.
b. Its effect on hitting the deepest variants. For
example you may have to have 50 african samples to
find the earliest offshoot branch of seqMRCA, this
will affect the MRCA calculation. One can actually
test this by randomly adding a known sample to the
data set and observe how the MRCA decreases with
added sample.
2. Sample site selection, did someone choose a site with a
handful of differences with chimpanzee. How does this
selection affect the external anchoring of mutation rates
3. Observations of potential variance based on differencese
of rates between humans, chimps and gorilla. To get at
this the cloud of low sampling and rate variance needs to
be removed. If the rate is removed how much are the rate
variances between gorilla and chimp be applicable. This
is a human factor because for example C/H distance may be
shorter than H/G arguing for slower evolution on the C
side of a branch and faster on H side, lowering MRCAs.
Each of these human errors can be randomly applied to an
randomly seleted MRCA inorder to give a humanizedMRCA that is
of publishable quality. this can be returned by the
MonteCarlo analysis and retested.
Based on analysis of the data set for which the similated
'humanization' process occurs I found several descrepancies
that I need to report, for the sake of keeping everyone
informed on molecular foul ups.
Xq21.2 (Yu et al) estimated human MRCA at 741 and popsize
about 14000. I thought this was a good paper, until I
started testing the observed differences in mutations C/H
and G/H, the authors should have provided C/G. It is
unfortunated because the C/G mutations are substantially
lowere than G/H and explains why the number of C/H
mutations is about 2/3rds that expected based on number
of G/H mutations. Based on C/G differences, almost all of
the rate variance is probably internal to Pan lineages
and then human rate is comparable to gorilla and
orangutan. This substantially lowers both Yu's MRCA and
the Population size Human factor. Did the human fairly
confidence the rate variance that might apply to the MRCA
and popsize determination. Rate determination in
chimpanzee local groups might determine if chimpanzees
are evolving significantly slower at this loci.
PDAH1. I was looking at their analysis. There rate of
mutations is consistent with other xlinked loci except the
fact that one arm of human haplotypes is evolving faster
(about 2 to 3 times) than the other arm. The slower evolving
arm has a rate that is consistent with other apes, but that
other arm is evolving about 3 times faster. This difference
could be due to a large population in a pre-constriction
period or it could be due to mutation rate variance. Its not
clear, but if the rate variance is applied to the faster
moving arm their MRCA and populations size (1.78 and 27 k
inds) drops to 1/2 that value. Human factor, did the humans
consider that intraspecific rate variation might occur.
Xq21.1. I always have something nice to say about Paabo. This
loci appears not to be evolving in an anomolous or
roller coaster fashion. It is probably most
representative of xlinked, although the sample size is
small (for example, I would like to have seen about 5
more samples from the biaka based on the samples they
presented from the biaka) Human Factor Is this MRCA
potentially deeper, more sample from biaka and closely
related peoples might reveal a deeper seqMRCA.
Xq22.2 factor IX locus. This locus suffers from a low number
of human substitutions, there appears to be balance
between chimp human, Orangutan not gorilla used as
outgroup so rate variation in the local group is hard to
survey. All in all this is not a bad study except that
when one has a small number of variant position one
probably needs to sample heavily where diversity is
deepest for the most reliable average of depth of
diversity. Human factor, inadequate sample of humans
I12rg. 9 difference between chimp and human, 29 gorilla and
human chimp and gorilla pairwise unknown. Means one mutation
for every 1.3 million years in the C/H lines, but 0 are found
in humans in a sample of 10, lol. Get more sample.
PLP. Get More sample, C/H difference to low, large percentage
error with 9 differences.
HPRT. Get More sample, C/H differnce marginal but C/H,
Orangutan/H are proportional to expected ages.
GK. Get More sample. C/H O/H are not proportional, There could
be rate variation at this loci.
Ids. Get More sample. C/H differences(5) has high
precentage variation
dmd7- I haven't got around to these yet
dmd44- I haven't got around to these yet. This is also Yu's
work. Yu collected both dmd 7 and dmd 44 variance, but decided
that dmd 7 MRCA (250 kya) wasn't but dmd44 at about 1 my was
representative, so . . . . . . .
What continues to suprise me is that these studies have a
number of statistics that tell them the sky is up and the
ground is down, and yet they still can't determine their ass
from their elbow. The Xq21.2 rate variation oversight was a
major goof. The conclusions of the paper were determinant on a
given rate and that rate can be shown to be significantly in
error. Also surprising is the continued use of small sample
size and poor outgroups sampling. Gorillas are included or
not, Orangutans or not, bonobos infrequently. It seems to me
that if there is a potency for rate variation, sequencing of
bonobo versus most divergent greater chimp would reveal if
there is something going on in chimp that would give pause to
modify the rate in humans.
Also surprising the number of differences between C/H and
J/H are frequently buried in papers, or often not determined.
Yu's most recent work, which includes some sample of X
chromosome (as well as many autosomals) 50 or so 500 nt
chunks from around the human genome, but no chimp outgroup.
Another surprise. Author will say one thing in one paper and
say a completely different thing in another without ever
questioning the validity of the first or newest paper. For
example, harris and hey propose a larger population size in
PDAH1 paper, but then with the FiX paper show population size
is smaller, Distribution? rate variation? Did they go back
and postulate rate vatiation within human PDHA1 haplotypes?
Nope. Then Yu proposes that human were of recent african
origins but left early, and there is this deep 140 ky branch
in eurasia, eurasion lines could not be from a few african
lines. Next study he shows that africans have 3 times the
diversity of eurasians, and that eurasians could be from a
few african lines.
I don't think Monte Carlo can factor in every human
behavior and bias, but I certainly hope it factors in
enough 'jockeying' of MRCAs in a random fashion to make it
of some use.
How does one distinguish whether a locus is underselection or
there has been a change in the mutation rate at the locus. I
wonder if a test for selection will wack the tester upside the
head and say 'hey look stupid, you've got rate variance at
this locus'.
. .
I have taken a rather, in hindsite painful task.
I have created a program that runs monte-carlo similuations.
Currently I have 12 xlinked popstudies that can be compared
with randomly generated sets based on different assumptions.
The initial runs gave pleasing results however a couple of
issues have bothered me.
Since I figured I had 12 individual author bias would be drawn
out by averages. But that the results of a couple of these
Il2rg and Ids loci would never be hit even by monte carlo.
This is problematic since Monte Carlo analysis could be
imployed to look at specific deviations, such as a particular
MRCA that is say in the 99.9% of ages. IOW I might have 10
loci that fit perfectly and have 2 that are so far in left
feild . . . .
All was going well until I ask the basic question. What is it
that we need to get to know exactly when an MRCA was.
1. The mutation rate (exactly)
2. The precise average of substitutions off the seqMRCA in the
current population.
I have just finished one Loci and I have found all kinds of
things that cause worry.
Plp locus 2 variant positions in humans 50% of sample have 1
variant derivative while 50% have seqMRCA. The difference
between the seqMRCA is 5 to chimpanzee, 10 for gorilla.
The binomial distribution on mutation rates based on 5
sequence differences ranges from .031 to 2.24 times the
calculated rate. For 10 it is a little better. The 96%
confidence range for relative rate differences between chimp
and gorilla is from gorilla evolving 5.5 times that off chimp
to 0.45. Thus these rates might apply to an evolutionary
segment like humans despite the calculated mutation rate.
This basic analysis has determined that a low number of
mutations between chimp and humans gets refactored into the
analysis on several levels. The variants at the loci may not
be reponding to the observed rate of mutation but a real rate
that is faster or slower than would be inferred from the
chimp/human absolute differences if those differences are
small. If there are 5 differences between chimp and human we
assume for instance that each difference represents
3.4 million years, and that since each individuals has 0.5
derivatives then the MRCA is 1.2 million years. But the
variance of potential differences of chimp and gorilla rates
could mean that the difference represents a few hundred
thousand or millions of years.
The basic bottom line. In order for MonteCarlo to be able to
compare human Derived MRCA versus the MRCAs expected in an
idealized model the human factor needs to be factored.
This factor can be broken down
4. Variance in sample size
a. Its effect in predicting the depth of mutations at a
site, and thus range of variation that could explain
observed values.
b. Its effect on hitting the deepest variants. For
example you may have to have 50 african samples to
find the earliest offshoot branch of seqMRCA, this
will affect the MRCA calculation. One can actually
test this by randomly adding a known sample to the
data set and observe how the MRCA decreases with
added sample.
2. Sample site selection, did someone choose a site with a
handful of differences with chimpanzee. How does this
selection affect the external anchoring of mutation rates
3. Observations of potential variance based on differencese
of rates between humans, chimps and gorilla. To get at
this the cloud of low sampling and rate variance needs to
be removed. If the rate is removed how much are the rate
variances between gorilla and chimp be applicable. This
is a human factor because for example C/H distance may be
shorter than H/G arguing for slower evolution on the C
side of a branch and faster on H side, lowering MRCAs.
Each of these human errors can be randomly applied to an
randomly seleted MRCA inorder to give a humanizedMRCA that is
of publishable quality. this can be returned by the
MonteCarlo analysis and retested.
Based on analysis of the data set for which the similated
'humanization' process occurs I found several descrepancies
that I need to report, for the sake of keeping everyone
informed on molecular foul ups.
Xq21.2 (Yu et al) estimated human MRCA at 741 and popsize
about 14000. I thought this was a good paper, until I
started testing the observed differences in mutations C/H
and G/H, the authors should have provided C/G. It is
unfortunated because the C/G mutations are substantially
lowere than G/H and explains why the number of C/H
mutations is about 2/3rds that expected based on number
of G/H mutations. Based on C/G differences, almost all of
the rate variance is probably internal to Pan lineages
and then human rate is comparable to gorilla and
orangutan. This substantially lowers both Yu's MRCA and
the Population size Human factor. Did the human fairly
confidence the rate variance that might apply to the MRCA
and popsize determination. Rate determination in
chimpanzee local groups might determine if chimpanzees
are evolving significantly slower at this loci.
PDAH1. I was looking at their analysis. There rate of
mutations is consistent with other xlinked loci except the
fact that one arm of human haplotypes is evolving faster
(about 2 to 3 times) than the other arm. The slower evolving
arm has a rate that is consistent with other apes, but that
other arm is evolving about 3 times faster. This difference
could be due to a large population in a pre-constriction
period or it could be due to mutation rate variance. Its not
clear, but if the rate variance is applied to the faster
moving arm their MRCA and populations size (1.78 and 27 k
inds) drops to 1/2 that value. Human factor, did the humans
consider that intraspecific rate variation might occur.
Xq21.1. I always have something nice to say about Paabo. This
loci appears not to be evolving in an anomolous or
roller coaster fashion. It is probably most
representative of xlinked, although the sample size is
small (for example, I would like to have seen about 5
more samples from the biaka based on the samples they
presented from the biaka) Human Factor Is this MRCA
potentially deeper, more sample from biaka and closely
related peoples might reveal a deeper seqMRCA.
Xq22.2 factor IX locus. This locus suffers from a low number
of human substitutions, there appears to be balance
between chimp human, Orangutan not gorilla used as
outgroup so rate variation in the local group is hard to
survey. All in all this is not a bad study except that
when one has a small number of variant position one
probably needs to sample heavily where diversity is
deepest for the most reliable average of depth of
diversity. Human factor, inadequate sample of humans
I12rg. 9 difference between chimp and human, 29 gorilla and
human chimp and gorilla pairwise unknown. Means one mutation
for every 1.3 million years in the C/H lines, but 0 are found
in humans in a sample of 10, lol. Get more sample.
PLP. Get More sample, C/H difference to low, large percentage
error with 9 differences.
HPRT. Get More sample, C/H differnce marginal but C/H,
Orangutan/H are proportional to expected ages.
GK. Get More sample. C/H O/H are not proportional, There could
be rate variation at this loci.
Ids. Get More sample. C/H differences(5) has high
precentage variation
dmd7- I haven't got around to these yet
dmd44- I haven't got around to these yet. This is also Yu's
work. Yu collected both dmd 7 and dmd 44 variance, but decided
that dmd 7 MRCA (250 kya) wasn't but dmd44 at about 1 my was
representative, so . . . . . . .
What continues to suprise me is that these studies have a
number of statistics that tell them the sky is up and the
ground is down, and yet they still can't determine their ass
from their elbow. The Xq21.2 rate variation oversight was a
major goof. The conclusions of the paper were determinant on a
given rate and that rate can be shown to be significantly in
error. Also surprising is the continued use of small sample
size and poor outgroups sampling. Gorillas are included or
not, Orangutans or not, bonobos infrequently. It seems to me
that if there is a potency for rate variation, sequencing of
bonobo versus most divergent greater chimp would reveal if
there is something going on in chimp that would give pause to
modify the rate in humans.
Also surprising the number of differences between C/H and
J/H are frequently buried in papers, or often not determined.
Yu's most recent work, which includes some sample of X
chromosome (as well as many autosomals) 50 or so 500 nt
chunks from around the human genome, but no chimp outgroup.
Another surprise. Author will say one thing in one paper and
say a completely different thing in another without ever
questioning the validity of the first or newest paper. For
example, harris and hey propose a larger population size in
PDAH1 paper, but then with the FiX paper show population size
is smaller, Distribution? rate variation? Did they go back
and postulate rate vatiation within human PDHA1 haplotypes?
Nope. Then Yu proposes that human were of recent african
origins but left early, and there is this deep 140 ky branch
in eurasia, eurasion lines could not be from a few african
lines. Next study he shows that africans have 3 times the
diversity of eurasians, and that eurasians could be from a
few african lines.
I don't think Monte Carlo can factor in every human
behavior and bias, but I certainly hope it factors in
enough 'jockeying' of MRCAs in a random fashion to make it
of some use.
How does one distinguish whether a locus is underselection or
there has been a change in the mutation rate at the locus. I
wonder if a test for selection will wack the tester upside the
head and say 'hey look stupid, you've got rate variance at
this locus'.