Rigorous multi-locus association mapping in genome-wide case control data is considered impractical due to computational and statistical limitations. We developed a method to efficiently apply a standard multi-locus test for synergy between SNPs. Our method overcomes computational limitations by employing a rapid, approximate sieve to drastically trim the universe of SNP combinations. The main idea is that we can dramatically cut computational costs if we only seek interactions that are strong enough to be statistically significant. We present our
implementation of this method (SIXPAC) that performs an order of magnitude fewer tests than an exhaustive search, but guarantees the inclusion of statistically significant effects in this subset. This reduction is achieved by a novel randomization technique which samples small groups of cases and highlights combinations of alleles jointly carried by all members of such a group. We applied this methodology to 7 common diseases in the WTCCC dataset and found synergy (p<10-12p<10-12) between several pairs of loci which are marginally unassociated.
Back to Workshop IV: Coancestry, Association, and Population Genomics