Investigating Synthetic Associations

Question

[Ryan D]

We are investigating the possibility of synthetic associations. We would like to identify coding variants in LD with GWAS identified variants using publicly available resources and are willing to look over long distances for them. As Kai Wang pointed out, r2 may be a meaningless measure for low frequency variants. Instead one may wish to look at measures of D-prime. Even D-prime of public samples (e.g. HapMap / 1KG) may not reflect long-distance LD present in a disease population. Given this stack of caveats, how would the BioStar community pursue such a question? We do not have complete sequencing data on these individuals so only those samples in public databases would be fair game.

Put on your thinking caps.

[Larry_Parnell]

I am curious as to why you wish to limit/focus this search on coding variants. Is this because you have exome sequence data available? Or are you hedging that coding variants, either synonymous or non-synonymous, will be the drivers of disease risk/disease phenotype?

[Larry_Parnell]

I am curious as to why you wish to limit/focus this search on coding variants. Is this because you have exome sequence data available? Or are you edging that coding variants, either synonymous or non-synonymous, will be the drivers of disease risk/disease phenotype?

[Ryan D]

Both. Sequence is coming, but we cannot get sequence data for our entire sample size. We were hoping to test whether or not coding variants exert "action at a distance" which leads to the GWAS signal. So for a GWAS significant association, are there one or more rare variants with some LD (within a disease population or otherwise) which might drive the signal?

Answer 1

[Larry_Parnell]

OK, coding variants could rule for your phenotypes. On the other hand, most disease risk variants arriving from GWAS and other types of studies are pointing to alterations in transcription, eQTL type stuff. It sounds like you need the opinion of a geneticist and for that you ought to contact my colleague Chao Lai: chao dot lai at tufts dot edu. He has given thought to this topic.

Not knowing how you define "long distance LD," it is hard to know where and by how much D' will fail or give values of low confidence. An alternative is consider boundaries defined by recombination hotspots, such as those described by Gil McVean et al.

So, three topics for you: 1) eQTL, 2) contact the geneticist, 3) recombination hotspots. One or more of these is likely applicable to your situation.