Comments on: EDAR again

By: Fly

Fly — Sat, 05 Jul 2008 15:20:31 +0000

gc: "1) SNPs are a priori going to be of small effect as they change only one bp. CNVs, by contrast, are changing kilobases or megabases at a time (i.e. entire genes or exons)."

Relevant links:
http://microarraybulletin.com/community/article.php?p=284&page=1
http://www.bio-medicine.org/medicine-news/Genetics-Of-Mental-Retardation-13248-1/

5% of the retardation cases are likely due to recent CNV's in brain genes. As far as I know all races are equally likely to experience such hotspot CNV's. These retardation CNV's are so harmful that they are rapidly eliminated by selection. I doubt they are the source of genetic group differences.

Other CNV's with less drastic effects are likely common. It would be interesting to compare the hotspot distribution differences between racial groups. Looking at CNV hotspots might make identifying CNV variation more efficient.

SNP's in regulatory DNA could significantly affect gene expression. Since there is twice as much conserved regulatory DNA as coding DNA in the human genome there are a lot of potential variants in the human population.

By: ben g

ben g — Sat, 05 Jul 2008 14:37:11 +0000

All the evidence suggests that environmental effects on a person's IQ wane as he ages.

elo basically took the words out of my mouth here,
but let me add that your statement needs to be clarified: what the evidence actually suggests is that the effect of the *home environment* *within a given group* diminishes with age.

what are the implications of that?

a) we must look at non-home environment effects as possible environmental causes of difference both within and between groups.

b) the home environment could serve as a cause of the difference between two groups, even if IQs regressed to the mean with age in both, because the home environment could plausibly be an early link in a causal chain which includes group differences (such as in socialization) later in life.

By: elo

elo — Sat, 05 Jul 2008 13:15:44 +0000

So, if the b-w IQ gap were due to differences in socialization, the gap between blacks and whites would be largest among children and smallest among adults.
What you have described is the opposite of the pattern you would expect to see if the differences were due to socialization.

By: Marc

Marc — Sat, 05 Jul 2008 12:29:22 +0000

I'm not willing to accept the hereditarian predictions in regards to the B-W gap at the time being because the most credible environmental hypothesis (which I observed even before I knew what IQ was), that blacks and whites are socialized in different cultural directions as they grow older, has not even been tested on a color blind basis, let alone between groups.

All the evidence suggests that environmental effects on a person's IQ wane as he ages. So, if the b-w IQ gap were due to differences in socialization, the gap between blacks and whites would be largest among children and smallest among adults. This isn't the pattern that we see.

By: ben g

ben g — Sat, 05 Jul 2008 10:17:47 +0000

I agree that's a good summary Rob. I think it's worth noting, though, that Flynn admits, for example, that Asians likely have innately superior viso-spatial abilities (Flynn 1980).

Since the brain has both centralized and modular aspects it stands to reason that populations will differ to some degree in both ways. For example, it might turn out that certain groups have genes that make them more susceptible to bad environments, or that other groups tend towards certain learning styles.

I'm not willing to accept the hereditarian predictions in regards to the B-W gap at the time being because the most credible environmental hypothesis (which I observed even before I knew what IQ was), that blacks and whites are socialized in different cultural directions as they grow older, has not even been tested on a color blind basis, let alone between groups.

By: gc

gc — Sat, 05 Jul 2008 08:21:30 +0000

Many people take the "no significant population differences" position, You may not, but it is quite common. In that case finding even just one allele that varies by population destroys the position. To keep no phenotypic difference, one would have to postulate even more genetic differences in the opposite direction. To maintain minimum genetic difference, one would have to accept some phenotypic differences.

Even if those individual variants don't hold up, this is exactly right. Good summary, Rob!

Moreover, it's extremely unlikely to get *exactly* the same phenotypic effects from different combinations of variants. Even given the (unsupported) hypothesis of extremely strong stabilizing selection on trait IQ, pleiotropic effects will mean that other traits start to diverge -- which is particularly significant for brain related genes.

The thing is, with traits like height, we *know* there's a significant amount of genetically-controlled variation from other data sources. So the initial SNP studies are just the start of sussing out the genetic architecture.

Similar considerations hold with IQ. It's just a matter of turning the crank at this point. It's going to be truly a thing of beauty when psychometrics -- "the one aspect of psychology which can be studied like a physical science" (sez the OG A. Jensen) is linked up with genomics, which has quickly become one of the most quantitatively demanding disciplines around (esp. in terms of CS/statistics).

Serious genomics requires a command of CS, stats, and basic genetics at a minimum. It is greatly aided by knowledge of the other two quantitative areas of bio: structural biology and population genetics. Among other things, this means that the people doing cutting edge bio research are increasingly nonconformist male math types, rather than your canonical obedient female benchtop biologist. Lots of factors gathering for a perfect storm.

By: razib

razib — Fri, 04 Jul 2008 22:11:12 +0000

Oops, thought the deleted troll comment was mine. Though that does make me a 'tard.

tardishness is contingent upon priors, so no worries :-) in any case, more impetuous than tardish.

By: Rob

Rob — Fri, 04 Jul 2008 21:41:52 +0000

Oops, thought the deleted troll comment was mine. Though that does make me a 'tard.

Ben, yes, probably hundreds or thousands variantions count. But honestly, would you have stated that sum total of the first three or four genes with frequent alleles affecting IQ would show population differences in the way that they actually do? That various "race realists" would have greatly strengthens their position. Their theory made better predictions.

Many people take the "no significant population differences" position, You may not, but it is quite common. In that case finding even just one allele that varies by population destroys the position. To keep no phenotypic difference, one would have to postulate even more genetic differences in the opposite direction. To maintain minimum genetic difference, one would have to accept some phenotypic differences. As intelligence and lifespan are widely considered significant, that argument is now greatly complicated.

Also, the SSADH gene has been under reasonably strong selection, and its distribution fairly closely matches either ASPM or microcephelin(too tired to find a link) There is no good reason to think that other alleles affecting cognitive ability or lifespan were invisible to selection.

By: Rob

Rob — Fri, 04 Jul 2008 21:21:35 +0000

In what way was my comment trolling?

By: ben g

ben g — Fri, 04 Jul 2008 14:27:09 +0000

from the linked paper: “The effect is small, with each allele having an effect size translating to about 1.5 IQ points.” so if this were the *only* gene effecting IQ, it would still account for less than an IQ point of difference on average.

By: ben g

ben g — Fri, 04 Jul 2008 14:26:05 +0000

So in addition to finding an allele that accounts for a portion of racial IQ differences, it also accounts for a portion of racial differences in both lifespan and declining cognitive ability with age.

not necessarily. there will be many more genes to come which affect the variance in these things.

By: Rob

Rob — Fri, 04 Jul 2008 10:10:27 +0000

SSADH is another gene associated with IQ that varies between races. The C allele version is more is a more efficient enzyme, and is associated with higher IQ .

Additionally, The C allele is associated with living longer. Popular article , Abstract Abstract

From the hapmap
population C allele
white 0.708
Chinese 0.9
Japanese 0.878
Yoruban 0.558

And from ensembl,
White 0.683
Chinese 0.867
Japanese 0.807
Yoruban 0.525

The genotypes deviated from Hardy-Weinberg. This could be current selection noise. As the two datasets had slightly different numbers, calculations from allele frequency and genotype frequency give these minimum and maximum IQ differences relative to whites:

pop max min
Chinese 0.576 0.5505
Japanese 0.51 0.3705
Yoruba -0.4755 -0.45

So in addition to finding an allele that accounts for a portion of racial IQ differences, it also accounts for a portion of racial differences in both lifespan and declining cognitive ability with age.

By: Marc

Marc — Thu, 03 Jul 2008 10:19:41 +0000

since a tally right now is statistically meaningless, i think the more notable story is that, on a given gene, there seems to be no consistent correlation between the IQ benefiting SNPs on the geographic level.

Why is this notable?

By: ben g

ben g — Thu, 03 Jul 2008 09:24:19 +0000

thanks for the link. half sigma's analysis does come out with a tally that favors whites/asians over blacks (and notably, whites over asians a bit). statistically speaking, though, a tally this early in the game is meaningless. dozens more genes will keep coming in, and they'll have effects as big, if not bigger than these.

i was following the primary election pretty intensely earlier this year. when the results would begin to come in for a state with less than 1% of the total vote in yet, it said nothing really about the final outcome (i especially like this analogy because you can think of counties as corresponding to individual genes.. individual counties report their poll tallies at different times, and you can't extrapolate from one county to another because of demographic variation) only around 40% did things start to stabilize, and the winner of a given state would become clear.

since a tally right now is statistically meaningless, i think the more notable story is that, on a given gene, there seems to be no consistent correlation between the IQ benefiting SNPs on the geographic level.

By: Marc

Marc — Thu, 03 Jul 2008 08:41:33 +0000

Assuming that a couple of the IQ genes have shown evidence of selection (i'll take your word on this), how does that...
...b) imply that the overall distribution of IQ genes favors a given population a great deal? of the few IQ genes (about 3 i can remember) that I've seen demographically analyzed, one of them favors blacks significantly, another favors whites, and another favors asians. so to claim that there is evidence that IQ genes geographically correlate with each other is preemptive based on the currently available genetic evidence.

Can you cite this? Folks did some number crunching over at Half Sigma's blog a while back and I remember that the majority of alleles coferred an advantage to whites or Asians. One allele (rs2619538) in DTNBP1 conferred a significant advange to Nigerians, but it was the only one of the seven on that gene associated with intelligence to do so, and the net result of that gene was still an advantage to whites and Asians. Half Sigma has a nifty graph here. You can read some of the number crunching in the comments of this post. I remember it being easier to follow the first time I read it, but perhaps that's due more to my lack of coffee right now than anything. :) If you've seen any other discussions on this sort of thing I'd be glad to see them.

Having said that, you're right that we've only found a few genes and I don't know if any of them have even been reproduced, so it's too early to say there is proof or strong direct evidence for a genetic basis for the B-W gap. I still think that the indirect evidence strongly supports the hereditarian position, however.

By: troll

troll — Thu, 03 Jul 2008 00:30:41 +0000

[IP address of troll: 67.135.15.12]

http://network-tools.com/default.asp?prog=trace&host=67.135.15.12

http://www.google.com/search?q=67.135.15.12&ie=utf-8&oe=utf-8&aq=t&rls=org.mozilla:en-US:official&client=firefox-a

By: ben g

ben g — Wed, 02 Jul 2008 20:37:54 +0000

Sure, one can conceive of theoretical scenarios in which we observe large behavioral differences (1SD is a *big* gap) but the underlying genetic variants totally wash out. That's highly unlikely for a number of reasons, not least because we're seeing selection on several of those neurological genes.

Assuming that a couple of the IQ genes have shown evidence of selection (i'll take your word on this), how does that
a) imply that the majority of other (as of yet) undiscovered IQ genes underwent selection?
b) imply that the overall distribution of IQ genes favors a given population a great deal? of the few IQ genes (about 3 i can remember) that I've seen demographically analyzed, one of them favors blacks significantly, another favors whites, and another favors asians. so to claim that there is evidence that IQ genes geographically correlate with each other is preemptive based on the currently available genetic evidence.

This gets at a larger point -- any individual argument re: brain imaging, geno/pheno, selection, psychometrics, etc. can potentially be critiqued if it's taken only on its lonesome.

That's a cowardly strategy and I don't think I've used it. I've offered[1] an alternative hypothesis for all the readers of GNXP to critique. You and others have been free this whole time to show how *it* doesn't fit into the big picture. If we have focused on a few narrow fields I think it's because the two proposed hypotheses (and we) agree so much on many of the core premises. Only a few lines of evidence are really able to distinguishing which of us is right about our differences (primarily studies related to peer groups and genomics)

...

technological advances are notoriously hard to predict. you surely keep up to date on many more of the relevant details than me, so i'll just try to keep an open mind on this issue (which is really tangential to our discussion anyways).

about the circlets, wouldn't that be social psychology (effect of groups on individuals and vice versa), not sociology (groups on groups)? i think sociology requires a damn good psychology, so i see it having useful theories as being a long way off, even if it suddenly became 100% empirical.

[1] Note that this hypothesis isn't original to me, but to various scholars I've read and tried to summarize/amalgamate.

By: gc

gc — Wed, 02 Jul 2008 19:43:05 +0000

I agree with all that. But note that most sociology today being unempirical doesn't mean that good, consilient sociology isn't part of the final answer to a lot of social-group level questions.

Oh, absolutely. I think some of the new generation are starting to bring such methods in through the back door -- particularly via MRI/fMRI.

That is, MRI isn't as controversial as genetics, in the sense that there isn't a propaganda literature devoted to demonizing the people who claim that there's a link between brain states and behavior (whereas of course there is in genetics, w/ mass-marketed Lysenkoism like "Not in our Genes").

So my prediction is that biology will increasingly start to come into sociology through the backdoor once we have cheap portable circlet fMRI with wireless (i.e. portable, 4D measurements of brain states). Idea is that you have a transparent, lightweight band with embedded solid state circuitry. Meant to wrap around the head and constantly transmit brain states back to a base station. This is the tech that will let you do truly quantitative microsociology -- videotaping encounters and correlating actions & reactions with brain states.

The fact that it's a circlet will allow people to potentially forget it's there and act "normally".

By: gc

gc — Wed, 02 Jul 2008 19:33:49 +0000

ben g:

(minor note, but i prefer the term "nurturist" to "environmentalist" as the former is less ambiguous than the latter)

1) It's possible that these genes keep coming out, showing signficicant *meaningful* variation, and the final genetic tally is basically even

Sure, one can conceive of theoretical scenarios in which we observe large behavioral differences (1SD is a *big* gap) but the underlying genetic variants totally wash out. That's highly unlikely for a number of reasons, not least because we're seeing selection on several of those neurological genes. Similar selection signals on other genes (e.g. melanocytes) are already known to have driven groups apart in phenotype, not together.

This gets at a larger point -- any individual argument re: brain imaging, geno/pheno, selection, psychometrics, etc. can potentially be critiqued if it's taken only on its lonesome.

But when you sum the totality of evidence (I've outlined it above, ranging from the lowest level SNPs to the highest level transnational/cross-cultural aggregates), it's not reasonable to assume zero or even modest genetic contribution, especially when the nurturists have yet to demonstrate a single reproducible gap-closing intervention.

2) how they have, after much work, gotten about 10% of the genetic variance in height.

Oh yeah, but it's well known that's b/c they were looking for common SNP variants on mass produced Affy 6.0 chips, and common SNPs are probably the least predictive variant (though technologically easiest to assay en-masse).

This is b/c:

1) SNPs are a priori going to be of small effect as they change only one bp. CNVs, by contrast, are changing kilobases or megabases at a time (i.e. entire genes or exons).

2) Moreover, *common* variants are unlikely to be of large effect just b/c their effects have been "debugged" in a lot of people and genetic backgrounds.

The CNV papers are new and I haven't compared the loci/datasets between studies (there are several papers to go through), but you're seeing significantly more explained variance -- and it's basically an orthogonal set of predictors to SNPs (that is, the early CNV map guys looked to see whether SNPs could proxy CNVs, but found relatively modest correlation between the two). Also, CNV discovery is still well underway relative to SNP discovery (which is relatively mature), so expect feature counts to follow SNP chips and rise dramatically over the next few years.

Of course, CNVs are just a stopgap. Th endgame is resequencing, which will give 3000X as much data as the current SNP chips (3 billion vs. 1 million data points per person). The conventional wisdom is that most of those 3 billion bp are constant from person to person, and that is true as a first approximation, but my feeling is that we're going to see a lot of wild stuff in which every region of the genome has at least *some* nonshared variation in say 1% of the people.

In other words, if you envision it as a matrix of characters with 3 billion columns and N rows, two things will likely happen given N even on the order of 1000:

1) every column has nonzero variance
2) no column can be completely discarded

Note that this is not how you'd actually deal with the data, b/c

a) different people's genomes will be of different sizes due to insertions, etc. so it's not a constant width matrix

b) in practice you'd want to featurize rather than dealing with a 3 billion column matrix; e.g. work with a derived feature matrix w/ 30k columns summarizing genetic variation, 100k columns for promoter regions, etc.

Just as an aside, I expect genomics to be driving hard drive sales for quite a while...750 MB compressed per genome times 6 billion genomes is a lot of data (about 4.5 exabytes!).

By: ben g

ben g — Wed, 02 Jul 2008 18:39:37 +0000

d’oh! “5, 7, and 10. I draw 8, 9, and 5. Our average score is identical – 11″ dividing by 2 instead of 3.. our average score would actually be 7 and 1/3