Women in engineering

There has been much serious discussion on this website and others on the relative dearth of women in the fields of science and engineering. This stereotype has become so pervasive in our culture that one bold engineering program, fed up with it all, took it upon themselves to prove, conclusively, that there are indeed women in their department. Naturally, the format chosen for their demonstration was the calendar. Preview [probably NSFW] below the fold:

Thrifty genotype hypothesis

For the population geneticist, diabetes mellitus has long presented an enigma. Here is a relatively frequent disease, often interfering with reproduction by virtue of an onset during the reproductive or even pre-reproductive years, with a well-defined genetic basis, perhaps as simple in many families as a single recessive or incompletely recessive gene. If the considerable frequency of the disease is of relatively long duration in the history of our species, how can this be accounted for in the face of obvious and strong genetic selection against the condition? If, on the other hand, this frequency is a relatively recent pheonomenon, what changes in the environment are responsible for the increase?

-James V. Neel, “Diabetes Mellitus: A ‘Thrify’ Genotype Rendered Detrimental by ‘Progress’?” (1962) [pdf]

The above quote is from the abstract of the highly influential paper by James Neel outlining the so-called “thrifty genotype hypothesis” for the prevalence of diabetes in modern populations. As this hypothesis is still widely cited (on this website, it has both praised and criticized), I present here my interpretation of the original paper, with comments as to how the hypothesis could be falsified or bosltered by the generation of unprecedented levels of population genetic data that we see today. I must note that this paper was written in 1962, and knowledge has certainly progressed since then. Some of the literature discussed by Neel or the paradigms he takes for granted are rather puzzling to me, and I may have missed some of the sublety of his arguments; readers are invited to peruse the .pdf linked above at their convenience and call me out on any mistakes.

I. The understanding of diabetes circa 1962

In 1962, “diabetes” was still considered more or less a single disease– it wasn’t until later that the current split between type I and type II diabetes was formalized. The adult-onset diabetes relevant to Neel is type II, so from now on, when I say “diabetes”, I will be referring to type II. Further, the notion of a “complex” disease was also absent– as is apparent from the quote above, Neel considered diabetes to be possibly caused by a single recessive allele, a situation subsequent research essentially ruled out.

In terms of the disease phenotype itself, Neel puts together a number of observations that suggest a certain advantage to the diabetic genotype. First, the children of diabetic mothers have increased birthwights as compared to the children of non-diabetic mothers. The children of non-diabetic mothers with diabetic fathers have higher than average birthweights, as well. Further, children who later develop diabtes tend to reach puberty slightly earlier, and thus could perhaps bear more children, than children who do not eventually develop diabetes. These observations, Neel suggests, indicate that the early diabetic phenotype is “thrifty”, in the sense that the children are particularly efficient in their use of the resources available to them.

Of course, the eventual diabetic phenotype consists of insensitivity to insulin and the inablility to properly process carbohydrates. Neel reconciles the apparent early “thriftiness” with this eventual insensitivity with a discussion of the physiology of diabetes. In this discussion, there are two major hormonal players who, while normally in equilibrium, are thrown out of balance in diabetes. The first hormone is insulin, which moves glucose from the blood to storage, and the second Neel refers to as “anti-insulins”, a class I can only assume refers to hormones like glucagon which release glucose into the blookstream.

In diabetics, then, Neel argues, there is an initial over-production of insulin, which accounts for the “thrifty” aspects of the genotype early on (the increased body weight and early menarche), which is then compensated by the stimulation of the “anti-insulins”. An inbalance results, and in adulthood this is manifested by insulin insensitivity[1]. The major question becomes, why has diabetes become so prevalent now?

II. The “thrifty genotype hypothesis”

Neel’s major insight in forming an answer to the above question is to note that “during the first 99 per cent or more of man’s life on earth, while he has existed as a hunter and gatherer, it was often feast or famine”. That is, there has been a marked change in environment in “modern” societies. He mentions three possible changes relevant to the control of the insulin/anti-insulin balance:

1. “Primitive” groups have less opportunities to overeat, have lower caloric intake, and greater physical activity than “modern” groups. This results in less stimulation of insulin, which in turn results in no over-stimulation of the anti-insulins.

2. The stress response in modern societies is less often followed by physical exertion than in primitive societies, which may disturb a “physiologic balance” established during human evolution.

3. The release of adrenaline in modern societies is also less often followed by physical exertion than in primitive societies. As adrenaline results in increased insulin production, this is an opportunity for the over-compensation of anti-insulins.

These last two are both similar in that they are involved with the stress response; indeed, Neel tentatively calls diabetes a “stress disease” along with peptic ulcers and hypertension!

In sum, the thrifty genotype hypothesis poses 1. that diabates results from a relative over-production of insulin, but more importantly 2. that “what we must now regard as an ‘over-production’ with unfortunate consequences was, at an earlier stage in man’s evolution, an asset in that is was an important energy conserving mechanism when food intake was irregular and obesity rare.”

III. The value of this hypothesis today

Now, I’ve noted the myriad assumptions made by Neel which are simply wrong– the assumption of a single gene being one of them[2]. I imagine the modern view on the physiology of diabetes is quite different than his as well. So what remains of this hypothesis?

I would argue that his key insight still remains valid– that, in population genetics, environment matters. Selection coefficients are not constant, and the way we alter our environment plays a large role in the selective forces exerted on us. More concretely, though, in terms of the genetic basis for diabetes, I can think of a couple predictions. First, any risk allele found for the disease will be ancestral– that is, a protective allele will have arisen recently. Second, the derived protective allele will have been under recent positive selection.

The prevalence of diabetes in different populations, assuming all have more or less the same diet, should also be negatively correlated with the time since switching from a hunter-gatherer lifestyle. This is the statement that is perhaps the most contentious– newcomers to the “modern”, high-carb diet certainly have high incidences of diabetes, but it’s impossible to tell whether this is due to the new availability of food or rather due to the content of the food itself. This may end up being a prediction that’s impossible to test, so my instinct is to stick with the genetic evidence. Of course, I’m a geneticist, so I would say that.

[1] I am
not at all familiar with modern reseatch into diabetes, and Neel’s views on all of this are likely a vast simplification or even entirely wrong. I don’t think, however, that this takes away from his later insights.

[2] An aside on Neel’s discussion of the genetics of diabetes. I found the following passage, under the heading “Some Eugenic Considerations”, to be interesing:

If the dietary and cultural conditions which elicit the relatively high frequency of diabetes in the Western World are destined to spread and persist over the entire globe, then, to the extent that modern medicine makes it possible for diabetics to propogate, it interferes with genetic evolution. But if, on the other hand, the mounting pressure of population numbers means an eventual decline in the standard of living with, in many parts of the world, a persistence or return to seasonal fluctuations in the availability of food, then efforts to preserve the diabetic genotype through this transient period of plenty are in the interest of mankind. Here is a striking illustration of the need for caution in approaching what at first glance seem to be “obvious” eugenic considerations!

Home labs

Coturnix made me notice that you can isolate DNA and run gels using legos and products from asian groceries. This got me thinking about what experiments I would run if I set this up. I think I could probably keep bugs of various sorts around and dissect out nervous systems. Could I select for certain traits in the right system? Could I discover anything new about biology in a DIY lab? There was a time when Rita Levi-Montalcini could make a huge contribution studying chicken embryos on a farm. What would you do with your home genetics lab? Are we at a point where the endeavor could only be a pastime or is there still something important to discover in our living room?

News to watch

This month’s Nature Genetics has a meeting report on the Human Genome Variation 2006 conference. This little bit caught my eye:

Andrew Clark and Neil Risch provided some exciting first glimpses into these large data sets, including some remarkable findings in population genetics. For instance, Andrew Clark examined global FST in different regions of the US and noted clear evidence of a gradient of allele frequencies that could partially be explained by the demographic history. He also noted some striking differences in heterozygosity and patterns of linkage disequilibrium between HapMap Caucasians and Ashkenazi subjects from an ongoing study.

Remember that natural selection can be inferred from linkage disequilibrium patterns. Andy Clark was never too keen on the selection for Ashkenazi intelligence thing; I wonder where these LD data are taking him.

Posted in Uncategorized

Direct measurement of the genetic contribution to the BW IQ gap

To follow-up on two older posts, here is a comment on the direct tests of the genetic contribution to the Black-White IQ gap that were proposed by David Rowe and Charles Murray. Each appears to be describing the same set of experiments. The aim of these experiments is to ascertain the relative contribution of genes versus environment to the Black-White IQ gap, or put another way they aim to measure the between-group heritability (BGH) of IQ.

The easiest way to explain what they are proposing is to steal a bit of text from a paper describing the analogous experiments for a different phenotype (lung cancer):

The explanation for the observed racial or ethnic variation remains to be determined. Unmeasured environmental variations, genetic differences, or both may be involved. Dissection of disparities among racial and ethnic groups is complicated by the strong correlation between the socioenvironmental and genetic factors that differentiate these groups, with few persons differentially classified.2,8 However, a number of approaches can be taken. One approach is based on the recognition of mixed continental ancestry among persons self-identified as African American or Latino. … Despite being genetically separable from whites, African Americans show a range of European ancestry that extends from nearly 0 percent to greater than 50 percent.9 Other studies have shown similar trends, with an average of about 20 percent European ancestry.10 Latinos are even more complex, comprising variable proportions of indigenous ancestry from three continental regions (Europe, the Americas, and Africa).11 Within these populations, individual ancestry can be estimated with the use of numerous ancestry-informative genetic markers; once established, this information can be used to examine correlations between the ancestry estimates and the trait of interest. [source]

Rowe and Murray each suggest examining the correlation between ancestry estimates (individual ancestry, or IA) and IQ. These experiments have not been done using the techniques of modern DNA genotyping, but they have been proposed for some time. Here’s my question: what would be the expected correlation between IA and IQ for a given BGH? A naive answer is that r = BGH. However, this is quite unlikely to be the case.

This question has been asked in a related context, so I’ll have to introduce more background to get us closer to an answer. While direct DNA genotyping has not been used to examine the IA-IQ correlation, other (less reliable) measures of IA have been used. Skin color is a prominent and notable example. Most recently, using data from the GSS, Lynn found a correlation of r=.17 between verbal IQ and skin color.[1] (I unknowingly replicated this finding in a previous post.) What is the implication of a correlation of skin-color and IQ of this magnitude to BGH? Jensen had previously considered this question.[2] In that context, Jensen estimates that the IA-IQ correlation should be around 0.5 and that the IQ-skin color correlation should be no more than about 0.2. Jensen’s 0.2 figure is based on an under-estimate of the IA-skin color correlation that is found using DNA markers and electronic measurement of skin color (decades later). (However, Lynn’s skin color data comes from self-estimates on a 5-point scale, and so Jensen’s numbers may be appropriate.) However, Jensen does not offer an explanation for how he arrives at a value of 0.5 for the IA-IQ correlation, and so it’s not clear what factors Jensen is taking into account. Jensen offers other reasons to downplay the importance of the IQ-skin color correlation as being informative about BGH. I’ll again steal text from the lung cancer review to explain the point:

Such analyses are not without caveats, however. Even within an apparently homogeneous admixed group, individual ancestry may remain correlated with environmental risk factors.8 This is most likely to be the case when ancestry is apparent or known, but less likely when it is cryptic. For example, in African Americans, skin pigment is correlated with the degree of European ancestry12 and may therefore lead to residual confounding. [source]

Getting back to the question: what would be the expected correlation between IA and IQ for a given BGH? I don’t know how to derive a formula to compute this directly, but it is easy enough to run simulations of the data. Jensen’s estimate of 0.5 is at the upper end of the values that I computed for BGH=100%. Why not r=1? The predominant reason is that IQ varies in the African American population for reasons in addition to variation in IA. It is difficult to know how much variation in IQ occurs at a given level of IA, but a lower bound estimate comes from the variation in IQ of siblings. Full siblings share the same level of IA (more than that, they are ~50% identical by descent), but show a substantial amount of variation in IQ. A common estimate I’ve seen is an average difference of 12 points among white siblings. Unrelated individuals with identical IA will vary at least this much (further correction for the lower overall variance in the Black population is needed). Factors of study design will also attenuate the IA-IQ correlation. While “Black” individuals may be found at all levels of IA, the actual population distribution of IA is clustered around 20% European / 80% West African ancestry; thus the range of IA measured will probably be restricted. Also, IQ scores have excellent but imperfect reliability, further attenuating the correlation. My attempts at simulating the IA-IQ correlation suggest that even if BGH=100%, the IA-IQ correlation might be as low as r=0.25. Jensen estimates BGH in the range of 50-75%, further reducing the IA-IQ correlation.There are other caveats, and possible way around these problems (MALD). From the lung-cancer review:

Another caveat is that an estimate of individual ancestry from the entire genome may be misleading if the racial or ethnic difference is due to one or a small number of genes.13 However, this is also an attractive scenario, since the same collection of markers could be used to pinpoint specific genetic locations involved in the difference (admixture mapping).10 In this case, the likelihood of residual confounding is reduced.13 [source]

MALD may be the best hope of circumventing the confounding of skin color with ancestry — the problem identified by Jensen, and later directed as criticism of Lynn’s conclusions[3]. If black-white skin color differences are mapped to a few loci of large effect (e.g. SLC24A5) then it should be possible to examine their effects in the MALD analysis. However, this all seems far from simple.

[1] Lynn, R. (2002). Skin color and intelligence in African Americans. Population and Environment, 23, 365-375.
[2] Jensen, A. R. (1973). Educability and Group Differences. London: Methuen.
[3] Hill, M. E. (2004). Skin Color and Intelligence in African Americans: A Reanalysis of Lynn’s Data. Population and Environment, 24, 209-214.


Here’s a figure that explains MALD in a case-control context:

Figure 1 | Detecting disease-associated gen
omic regions using mapping by admixture linkage disequilibrium. a | The strategy that is used to assess the ancestral origin of chromosomal segments in mapping by admixture linkage disequilibrium (MALD)7, 13, 15 . Genotyping MALD markers is used to assess parental ancestry across a single chromosome in multiple cases (individuals with the disease of interest) versus matched healthy controls. The region indicated by the star is derived more often from one of the parental populations only in the disease cases, indicating that this region contains a disease-susceptibility locus. In the controls, the same region has an equal probability of originating from either parental population. b | A theoretical example of how an admixture signal can be detected using the MALD method for a disease with a higher incidence in one parental population (population A). The proportion of ancestry from population A in multiple individuals (both with the disease (cases) and without the disease (controls)) is shown schematically for different positions on a single chromosome. An elevated ancestry proportion from population A in cases is evident at the peak (marked by an arrow), which indicates the involvement of the corresponding genomic region in the disease. The peak can be identified by the higher (or lower; not shown) level of ancestry that is seen in cases relative to the same region in controls, and/or relative to the remainder of the genome in cases (only the neighbouring chromosomal region is shown here). Part b is modified, with permission, from Ref. 13 © (2004) The University of Chicago Press.
© 2005 Nature Publishing Group


The Pig Men Cometh

“I wish there were pig-men. You get a few of those pig-men walking around, suddenly I’m looking a lot better.”

-George Constanza of Seinfeld

Of course, a world with pig men would be a less beauteous thing. Kind of what I thought when I read Steve’s most recent column on the rise of academic inequaliy, and ascendency of sub-mediocrity, in the Los Angeles school system.

Norm of reaction and Williams Syndrome


Despite the differences in upbringing, in both countries children with Williams syndrome were rated significantly higher in global sociability and their tendency to approach strangers than were their typically developing counterparts. But cultural expectations clearly influenced social behavior, since the sociability of normal American kids was on par with Japanese Williams syndrome kids, whose social behavior is considered out of bounds in their native country.

Norm of reaction “describes the pattern of phenotypic expression of a single genotype across a range of environments.” So it seems there is a weakness in this study: Americans (mostly of European ancestry) do not share the same genetic background as Japanese. So a better test would be Japanese American children with Williams Syndrome vs. Japanese.

Icelandic fire

This week’s Lancet has a profile of Kari Stefansson, CEO of DeCode Genetics. Regular readers have been exposed to much of the groundbreaking research done by the company, which has DNA samples from ~65% of the Icelandic population and a geneology that stretches back 1000 years. This is the group that published on the “fertility inversion“, alleles that cause an ethnic-group-specific risk for heart disease, and generally has been one of the few groups to have success finding alleles that contribute to succeptibility for complex diseases like prostate cancer or diabetes.

Could anyone with a dataset as good as the entire Icelandic population pull this off? Stefansson:

“No, no, no. The real advantage is just that we are the best scientists, alright? Don’t give me this bullshit about our advantage being the [Icelandic] population. Why do we have this population? Because we realised the importance of it.” Although the tone of his voice suggests that he is being slightly tongue in cheek, it is clear that he is serious about the underlying message.

His repsponse to a question we’ve often asked interviewees– any interest in your own genome sequence?

Stefansson says that he is not planning to follow Craig Venter’s example and have his own DNA sequenced. “My mother died at the age of 62. My father died at the age of 67. And therefore I have been very diligent about avoiding to learn anything about my own disease predisposition”, he says. “I want to die ignorant of my weaknesses.”

Posted in Uncategorized

Pinker on consciousness

Steven Pinker has an article in Time called “The Mystery of Consciousness”. An extract:

What remains is not one problem about consciousness but two, which the philosopher David Chalmers has dubbed the Easy Problem and the Hard Problem. Calling the first one easy is an in-joke: it is easy in the sense that curing cancer or sending someone to Mars is easy. That is, scientists more or less know what to look for, and with enough brainpower and funding, they would probably crack it in this century.

What exactly is the Easy Problem? It’s the one that Freud made famous, the difference between conscious and unconscious thoughts. Some kinds of information in the brain–such as the surfaces in front of you, your daydreams, your plans for the day, your pleasures and peeves–are conscious. You can ponder them, discuss them and let them guide your behavior. Other kinds, like the control of your heart rate, the rules that order the words as you speak and the sequence of muscle contractions that allow you to hold a pencil, are unconscious. They must be in the brain somewhere because you couldn’t walk and talk and see without them, but they are sealed off from your planning and reasoning circuits, and you can’t say a thing about them.

The Easy Problem, then, is to distinguish conscious from unconscious mental computation, identify its correlates in the brain and explain why it evolved.

The Hard Problem, on the other hand, is why it feels like something to have a conscious process going on in one’s head–why there is first-person, subjective experience. Not only does a green thing look different from a red thing, remind us of other green things and inspire us to say, “That’s green” (the Easy Problem), but it also actually looks green: it produces an experience of sheer greenness that isn’t reducible to anything else. As Louis Armstrong said in response to a request to define jazz, “When you got to ask what it is, you never get to know.”

The Hard Problem is explaining how subjective experience arises from neural computation. The problem is hard because no one knows what a solution might look like or even whether it is a genuine scientific problem in the first place. And not surprisingly, everyone agrees that the hard problem (if it is a problem) remains a mystery.