ASHG: Statistical Genomics and Beyond GWAS in Complex Disease

The second day of the American Society of Human Genetics Annual Meeting is drawing to a close; here’s a lowdown of what talks I’ve enjoyed today.

Remember, follow @lukejostins on Twitter if you want more up-to-the-minute details on the ASHG talks.

Statistical Genomics

It was good to finally get down to some more meaty statistical tests. A couple of people presented new twists on association tests; Dajiang Liu explored tests combining extreme phenotype sib-pairs and unrelated individuals for tracking down eQTLs, Gregory Kryukov used PolyPhen scores to squeeze significance out of associations for coding SNPs, and Terresa Ferreira showed that you could get extra power by including known GWAS hits as covariates when looking for new hits. None of these felt like anywhere near the final world on rare variant association, but it is a good sign that there is such a thriving research community working on the problem.

Bayesian analysis made a welcome contribution to the session. Clare Churchouse gave a Bayesian method for inferring individual ancestry; the test allows you to find which population each base of DNA came from. She can do this with >95% accuracy in European and African admixed individuals, and between 75-85% for individuals from an admixture of the more closely related Japanese and Chinese populations. If and when she extends this to a large number of potential populations, this could be very interesting for the personal genomics community.

David Balding gave an awesome talk, a version of his equally awesome paper in Nature Reviews Genetics on Bayesian association studies. There is a major contradiction in classical association studies; we assume that p-values are a meaningful measure of statistical support; which is only true if all sites are tested with equal power (clearly false). He showed how a proper Bayesian analysis gets rid of this problem by giving a proper measure of support (the posterior). He also talked about how a Bayesian approach can fix many of the problems with classical meta-analyses. It was, as I said, awesome.

Beyond GWAS in Complex Diseases

The other session I went to this morning was a set of Great Men (alas, all men, for the first time this conference) of Complex Disease talking about where to go, now that the low hanging fruit of common variants have been mopped up in GWAS.

Peter Donnelly gave an exciting account of all the hard work, clever tricks and technical wizardry that went into doing CNV association across the WTCCC samples; a story with a cruel end, and no real novel discoveries. As is now known to everyone, common CNVs play very little role in complex disease, and those that do are already tagged by SNPs.

Goncalo Abecasis gave a wide ranging talk about various types of post-GWAS work, ending with a discussion of sequencing association studies; he suggusted that a hybrid approach of high-coverage exon resequencing and low-coverage whole genome resequencing may be the way forward.

Jeffrey C. Barrett (my surprisingly youthful Ph.D supervisor, who nonetheless managed to fill the Great Men role with surprising finess) gave a talk looking at disease genetics from a multi-phenotype viewpoint. He showed how a multi-phenotype association test can find new associations in Inflammatory Bowel Disease data, and how interesting patterns emerge when we look at what risk genes are shared between different autoimmune diseases.

I came out of this session feeling quite refreshed; it is a common belief that disease association has hit a brick wall, but all the clever and experimental approaches that were discussed showed that there is still a lot a lot of lines of enquiry that are forging ahead. Of course, this being my field, I was already dimly aware of these approaches; however, having them all displayed together made them into a collective project, a joint endeavor.