Association graph reduction: a framework for efficient exploration of high-level interactions in complex disease etiology

Jeffrey Kilpatrick, Rice University

Photo of Jeffrey Kilpatrick

Advances in genotyping technology have led to overwhelming availability of data. In spite of dense marker panels and powerful association statistics, the causes of most phenotypes with complex etiology remain elusive, and even strong statistical effects often evade replication. One possible cause for such failure to confirm results is the action of epistatic interactions among causative factors, which may display fleeting association or no marginal effects whatsoever. Existing methods either tend to overlook factors exhibiting no marginal effect or are too computationally expensive to be used on a genome-wide scale. We present association graph reduction, a novel framework for the exploration and evaluation of causative factors of complex phenotypes. We demonstrate its efficacy and compare it to multifactor dimensionality reduction (MDR), a popular existing method, which exhaustively evaluates all possible interactions involving a specified number of factors. Results show that our method is as accurate as MDR and orders of magnitude faster.

Abstract Author(s): Jeff Kilpatrick and Luay Nakhleh