Would be interesting to compare the prediction power you can get with current loci to the maximum achievable one for each disease (based
on heritability, from e.g. twin studies etc.)

Or

Multiplying odds ratios together is valid only if the risks for each variant are independent; however, it seems that most common variants have relatively small interaction terms. From what I’ve seen, I’d expect the odds ratios calculated above to be slight overestimates, but probably pretty close guides.

I’m working on genetic models for explaining GWAS results, and
have found this entry very interesting.
May I ask how was the disease risk distribution you’ve plotted computed?
So my best guess is that for each of the 3 papers, you extracted for each SNP both its population frequency and it’s odd ratio for getting the disease.
So for example for chron’s disease, you’ve got 30 loci and you know their frequencies. Now you can sample genotype vectors with these 30 loci, and for each such genotype you need to compute the disease risk – so what is your assumption when computing this? do you just multiply the odds ratio, i.e. assume a multiplicative model? or are there any other assumptions? I guess without knowing the true genetic architecture and without further assumptions you cannot really know the distribution of disease risks. Any clarifications would be greatly appreciated.

Best,
Or

Clayton, D.G. (2009) Prediction and interaction in complex disease genetics: the experience in type 1 diabetes. PLoS Genetics, 5 (7), e1000540.