All the ASHG talks that I have had to do analysis for have now been given, so today I’ve managed to dedicate my full attention to the sessions. Also a good day for tweeting; I managed to live-tweet quite a few talks on @lukejostins, and the #ASHG2010 hashtag has been totally rammed.
Larry Parnell over a Varigenome has been putting his ASHG notes up, if you are still hungry for details. Daniel Macarthur has promised a post on the “Identifiability in the Era of Genome-Scale Research” session for Genomes Unzipped, and I saw him getting pretty worked up about Jim Evan’s talk on his twitter feed, so hopefully we’ll see something from him as soon as he’s done being dead of plague.
Two sessions today, “Statistical Analysis of Human Sequence Variation” and “Finding High-Risk Susceptibility Gene Variants”, seem to encapsulate the cutting-edge of disease gene association, and illuminate where disease genetics is heading in the immediate future.
As with last year, there was a number of next-gen association tests presented, trying to extend the GWAS testing toolkit to test for hard-to-detect variants, such as ones that are low frequency or small effect
Iuliana Ionita-Laza presented a new score test, and showed that it beat a few existing methods at detecting signal at IFIH1, and Michael Schmidt presented a way of accounting for uncertainty in sequencing association (though failed to compare it to general uncertainty-handling tests like Plink and SNPTEST). Via a poster, Kate Morley presented more results from her multinomial elastic net regression, taking two new IBD genes forward for replication, and successfully replicating one.
Ben Neale and Momiao Xiong both presented some novel tests for gene-wise association. Both use general methods for detecting deviation from the expected null frequency distribution; the C-alpha method for detecting deviation in binomial mixtures and a principal component approach, respectively. Both showed simulations showing that these can, in certain circumstances, seriously increase power.
This is all coming together, if slowly; last year, I predicted that this year would be the year of successfully rare variant association, but it never quite appeared. However, the methods are all quite mature now, and I would put money on there being a large, successfully replicated genome-wide study applying a gene-wise test for low-frequency/low-effect variants.
Slightly less mature, but perhaps more exciting, were the reports of family-based studies with large pedigrees, looking for rare variants of large effect (“large effect” seems to vary in meaning from “a Relative Risk of 5″ to “basically Mendelian”). A few of the talks (by Joan Bailey-Wilson and Ellen Wijsman) had no well-replicated results, but lots of very interesting methods, and informative failures.
Nicola Camp talked about the amazing Utah Population DataBase, who have collected 2.2M individuals from 3+ generation pedigrees, along with digital medical records. The ability to pick out the most severe, most well-powered and most Mendelian pedigrees for genotyping makes their linkage more successful than most.
There was a lovely story from Rose Yang, who talked about a study of four families with high incidence of chordoma. They found a solid linkage region on chromosome 6, which contained a nice candidate gene encoding the protein brachyury, which is overexpressed in chordoma tissue. However, sequencing the gene found no mutations that segregated with disease, and moving on to another 20 genes didn’t give anything else. Finally, they used array-CGH to scan for novel CNVs, and found a large insertion right in the middle of the brachyury gene in one of the 4 families (present in 6 affected and 0 unaffected individuals).
Anyway, if anyone needs me, I’ll be running an array-CGH chip on some of my disease pedigrees…