Greg Cochran pointed out something that I’d been considering about the MacArthur et al. paper: if the average human (OK, non-African human) has ~100 loss-of-function variants, then the standard deviation should be ~10. That’s because the distribution is presumably poisson, and variance = mean, and the square root of the of the variance (~100) is the standard deviation (~10). In plainer English there should be a substantial variation in the number of loss-of-function variants within a population, and across siblings. Though by definition these loss-of-function variants don’t kill you, in general there is the assumption that this class of mutants does exhibit some fitness drag (e.g., the fitness of a heterozygote for a variant which is lethal as a homozygote genotype may be ~0.90). A quick back of the envelope calculation implies to me that there is a 1 out of several hundreds of thousands probability that two siblings may exhibit a range of 60 loss-function-variants. But a 40 unit gap is more like a 1 out of one thousand chance.
This variance in mutational load has been the hobby-horse of intellectuals for a while now. Armand Leroi suggested that it correlated with beauty. Geoffrey Miller with intelligence. In the near future presumably we’ll get to see if there’s anything real in this. And obviously we don’t need to leave it to scientists. We’ll all know the summary statistics about own genomes, and probably be able to intuit rough patterns…if they exist.
Razib Khan’s degrees are in biochemistry and biology. He has blogged about genetics since 2002, previously worked in software development, is an Unz Foundation Junior Fellow and lives in the western US. He loves habaneros.
February 22nd, 2012 at 9:33 am
poisson
I don’t trust WP, so to verify: Poisson is used for mutations _____.
A) Usually, general case?
B) In studies like this, because they cast a wide net?
C) ?
February 22nd, 2012 at 1:58 pm
#1, low frequency event. my understanding is that usually in biology it *underestimates* variation. also, they might have the real distribution in their sample in the supps. didn’t check.
February 23rd, 2012 at 9:01 am
Do the calculations take into account the fact that the sibling rates are correlated? The distribution of a difference of independent Poissons is a Skeelen distribution which is probably close to a normal in your example. Siblings though should be correlated right?