Just out in prepublication at PNAS is a paper from Eric Lander’s lab, entitled, somewhat provocatively The mystery of missing heritability: Genetic interactions create phantom heritability. The authors suggest that certain types of gene-gene interactions could be causing us to underestimate how much of the heritability of complex traits has been uncovered by our genetic studies to date.

There has been an awful lot of talk about this research since Eric Lander talked about it at ASHG a few years ago, and the paper itself has generated quite a bit of discussion on- and off-line. Razib Khan reported on the paper last week, giving a good summary. He mentioned a press release about the paper issued by the advocacy organisation GeneWatch, which confuses the additive heritability discussed in this paper with the total heritability of diseases (a distinction explained below), and uses this to draw conclusions about how this result alters the promise of personal genomics. This just goes to show how much confusion there already is out there about this subject.

I have a more detailed post up on Genetic Inference about this paper, the strength of the argument, and what it means for the field. Here I am just going to pull out what I think are some important take-home points about this paper:

1) Broad sense heritabilities (the kind that are clinically important for e.g. risk prediction) have NOT been significantly overestimated The type of heritability we ultimately care about, the broad or total heritability, is how much total phenotypic variation is captured by genetics, or equivalently the correlation between identical twins in uncorrelated environments. The figure at the top of this post shows a plot that I made using Zuk et al’s equations, comparing true broad sense heritabilities, against what would be estimated based on twin studies (I have matched the colouring etc to Figure 1 of the paper). The twin study estimator of heritability is a robust estimator of total heritability for heritabilities less than 0.5. Above that, LP epistasis causes growing overestimation – it can make a 50% heritable trait look like a 65%, and 70% look like a 95%. It does not make weakly heritable traits look strongly heritable, just strongly heritable traits look very strongly heritable.

2) This paper is discussing additive heritability. This is a specific form of heritability that acts “simply” – half of it is passed on to offspring, siblings share an amount proportion to how related they are, and the genes that underlie it do not interact with each other. We do not know how much heritability acts like this, but various lines of evidence have made us think that it is a relatively good model, and most competing models have been incompatible with this evidence, or look contrived. What Zuk et al have done is produce a set of plausible, simple and non-contrived models (Limiting Pathway or LP models) that look pretty much indistinguishable from additivity using many of the tests we have run, but can act very differently in twin studies. Under these models, twin studies will overestimate the additive heritability (i.e. make us think that a larger proportion of heritability acts “simply”). The equivalent plot to the top of the page for estimating additive heritability, which you can see here, shows massive overestimation of additive heritability across the spectrum.

3) There is no real evidence that these LP models apply (and in fact there are still a few reasons to believe additivity could still broadly apply, see my other post for details). The issue is that we cannot conclusively rule these models (or models like these) out, and therefore the heritability explained by the genetic variants we have found so far is very uncertain.

4) This is important because our measures of “heritability explained” by the genetic variants we have found look at how much additive heritability is explained. These measures have in general told us that we have only explained a small proportion (generally < 25%) of additive heritability – but if in fact the heritability is largely not additive, but we are treating it like it is, we could in fact have explained a higher proportion of heritability than we believe. This would mean that the “missing heritability” is missing not because we have not found the right genetic risk factors, but because we have not found the right model to use. This could be good news: the genetic variants we have discovered could in fact be used to predict disease a lot better than they we can at the moment, if only we can find the right model to use them with.

[This is a guest post by Alex Kogan. Last week, Ed Yong at Not Exactly Rocket Science covered a paper positing an association between a genetic variant and an aspect of social behavior called prosociality. On Twitter, Daniel and Joe dismissed this study out of hand due to its small sample size (n = 23), leading Ed to update his post. Daniel and Joe were then contacted by Alex Kogan, the first author of the study in question. He kindly shared his data with us, and agreed to an exchange here on Genomes Unzipped. Our comments on the study are here; this is Alex’s reply.]

It’s a truism that resonates across science: Size matters when doing and interpreting the statistical (and practical) meaning of a study. But the size of what? Well, it’s quite a few things—all of which are very important in understanding what a study is ultimately telling us. One of the first numbers researchers focus on is the p-value. The p-value relies on a bit of counterintuitive logic: It represents the percentage of times you would get an effect as big as you got (or bigger) if there is really no effect in the general population. So we first assume that there is really no difference in some outcome between two groups across the general population (we call this the null hypothesis), and then we ask what are the chances of us finding the difference that we found (or bigger) given this assumption. If this percentage is low (many fields adopt a p = .05 standard, or a 5% chance that we’d get the effect we got or bigger if there is really no effect in the general population), then we can reject the initial idea that there is no difference in the general population. So what have we learned if the p-value is .05 or lower? That there is likely a difference in the general population—how big this difference is remains a mystery, however; the p-value never answers that question.
Continue reading ‘Guest post from Alex Kogan: Size and populations matter–let’s understand why’