I’ll be uploading a few of what I saw as the highlights from the AGBT meeting over the next week or so, as I go over my notes – you can also browse over Anthony Fejes‘ blog for live-blogging of many of the sessions. In no particular order, here are some of the tid-bits gleaned from Friday’s sessions.
Kari Stefansson gave an overview of some of the latest results coming out from deCODE Genetics. He argued that combining multiple genetic variants for common diseases can now give clinically useful results – e.g. their work on thyroid cancer (published today in Nature Genetics) shows that the 3.7% of individuals with the riskiest set of variants had a nearly six-fold greater risk of disease.
Stefansson also presented two fascinating pieces of unpublished data. Firstly, he discussed the use of large-scale family information to enable accurate inference of long-range haplotypes (basically, figuring out which chunks of chromosome you inherited from your father and which from your mother) – this approach was published late last year. The new data was the application of this technique to disease inheritance: deCODE has identified variants in type 2 diabetes and other diseases that confer totally different effects on risk depending on whether they were inherited from your mother or your father. This suggests the fickle hand of parental imprinting; if it turns out to be a more general mechanism this will greatly complicate the hunt for disease-causing variants.
The other interesting data-point comes from a large genome-wide association study of cognitive traits, performed via a web-based survey of thousands of Icelanders who had previously been genotyped by deCODE. Stefansson claims that this study has identified a genetic region that is significantly associated with a love of crossword puzzles. I’m not kidding. Although I can see the facetious headlines already, this type of association is not just frivolous – understanding the genetic architecture of any behavioural trait gives us a better understanding of the molecular basis of personality differences.
Another highlight was an excellent presentation by Cambridge University’s John Todd, who discussed recent data on the genetics of type 1 diabetes. Genome-wide association studies have already found 42 regions of the genome containing common variants associated with this disease, and Todd’s group has been performing deep sequencing on some of these genes to see if they also harbour rarer disease-causing variants. One gene (IFIH1) contains two rare variants with a very convincing association with type 1 diabetes risk; they are present in just 2.2% of the normal population, and are actually protective against disease (odds ratio ~0.5).
Both Todd and the following speaker, Baylor’s Richard Gibbs, took a very different view to Stefansson regarding the value of current common disease-associated variants for clinical practice. Todd pointed out that using current genetic markers we can identify 20% of children containing 70% of type 1 diabetes cases, but that’s still nowhere near good enough to guide clinical interventions. Gibbs was more forthright: he noted that he had been genotyped by a personal genomics company, and “the information was completely useless.”
Like Todd, Gibbs emphasised that rare variants almost certainly contribute a substantial fraction of genetic disease risk, and described several ongoing projects at Baylor to find these variants through high-throughout sequencing. I was very pleased to hear this, because in order for my prediction that 2009 will be the year of rare variants for common diseases to come convincingly true there will need to be a whole bunch of these variants identified and published over the next twelve months.
Also, a quick note on Complete Genomics: I met with the CEO and CSO of the company on Saturday regarding questions I had about both the technical and business facets of their plans. I’ll hopefully have a post up on this tomorrow.