Computational Methods to determine Epistatic Effects

Regents Of The University Of Michigan - Ann Arbor, Ann Arbor MI

Investigators

Abstract

Hanlon 0073785 The investigator designs effective algorithms to identify epistatic effects that contribute to a complex quantitative phenotype. The algorithms apply to datasets that, for each individual in a sample, identifies alleles at a fixed set of markers along with the value of the quanitative trait. The algorithms seek an optimum fit for the phenotype as a sum of epistatic effects. The approach is to perform repeated searches for an optimum fit in the space of all possible combinations of effects. The search utilizes random walk hill-climbing techniques. These hill-climbs are adaptive in the sense that certain parameters, which control the choice of steps in the random walks, evolve based on the outcomes of previous walks. It is widely accepted that the genetic components of most interesting quantitative traits (eg. blood pressure, height) involve interactions between many genes. So, computational methods to infer the genetic basis of quantitative traits must include an analysis of epistatic effects, i.e., instances where the choice of allele at two or more genetic markers influences the value of the quantitative trait in ways that are different from the sum of the effects of the individual alleles. In this project the investigator develops computational methods to identify these epistatic effects by a process of doing repeated searches through the space of possible effects looking for ones that explain a significant amount of the variation of the trait. These searches are self-educating in the sense that later searches learn from earlier searches. As the number of searches increases, the adaptive mechanism focuses the search on particularly strong epistatic effects.