A Recommendation Systems Approach for Detecting Epistasis in Genomic Signals

There are a variety of methods used to understand and interpret an organism's phenotype, the physical expression of one or more genes. Epistasis, the phenomenon of one mutation affecting the resulting quantitative or qualitative phenotype, is used to assess gene variation in an attempt to find a combination of single nucleotide polymorphisms (SNPs) that contribute to a certain phenotype. Since one SNP rarely completely describes an organism's phenotype, detecting these groups, or coalitions, of mutations without relying on an exponential number of numbers is one of the main challenges in this field. To alleviate these computational bottlenecks, we propose a neighborhood-based collaborative filtering approach by viewing this data with a recommender system formulation. As such, we are able to detect statistically significant higher order SNP interaction phenotypes related to muscle mice genomic variants.