GRAIL: Gene Relationships Across Implicated Loci

If you caught Soumya Raychaudhuri's seminar last week you heard a lot about the tool he developed at the broad called GRAIL - Gene Relationships Across Implicated Loci. You've got GWAS results and now you want to prioritize SNPs to follow up in replication or functional studies. Of course you're going to take your stellar hits at p<10e-8, but what about that fuzzy region between 10e-4 and 10e-8? Here's where a tool like GRAIL may come in handy.

In essence, you feed GRAIL a list of SNPs and it maps these SNPs to gene regions using LD. It then uses a simple text-mining algorithm to ascertain the degree of connectivity among the associated genes by looking at the similarity of vectors of words pulled from PubMed abstracts which mention your gene of interest. In their most recent paper they took a list of 370 GWAS hits, and narrowed this down to a list of 22 candidate SNPs to follow up.  And it turns out these SNPs replicated in an independent set at a much higher frequency than random SNPs from the subset of 370. In his talk, Soumya offered convincing evidence that using the results from GRAIL you have a much better shot at replicating associations than if you just looked at the p-value rankings alone.  After the talk he did mention that this approach has had mixed success depending on the phenotype. Here's the original GRAIL paper, and the OpenHelix blog has a nice 5-minute video screencast demonstration of GRAIL where they take a few SNPs from the GWAS catalog and run them through GRAIL.

GRAIL is a free web application (beta) available at the Broad's website below.

GRAIL: Gene Relationships Across Implicated Loci