Epistasis: Obstacle or advantage

Identification of genetic loci in complex traits has focused largely on one-dimensional genome scans to search for associations between single markers and the phenotype. There is mounting evidence that locus interactions, or epistasis, are a crucial component of the genetic architecture of biologically relevant traits. While discussion about the best way to identify models for complex traits is ongoing, recent work shows that fitting full models in exhaustive multi-locus genome scans can have higher power to detect epistatic loci than single-locus scans. This improvement comes despite a much larger multiple testing alpha-adjustment cost due to a dramatically larger number of tests that are performed. We demonstrate, both theoretically and via simulation in an association mapping context, that power to detect two loci when fitting full models is often larger when these loci act epistatically than when they act additively. Additionally, we show that in cases of molecular epistasis the power for single locus detection may be improved compared to the additive model. This indicates that the impression that epistasis is a nuisance factor that reduces power for locus detection may be false. Our brief exploration of a two step model selection procedure shows that identifying the true model is difficult, although this difficulty is certainly not exacerbated by the presence of epistasis and in some cases epistasis may aid in model selection. As technology becomes more cost effective, the amount and scale of data available for answering fundamental questions about the underlying genetic contribution to phenotypic outcomes of interest has dramatically increased. Genetic markers, particularly biallelic single nucleotide polymorphisms (SNPs), are being developed for a wide variety of organisms et al. 2005), and current SNP discovery techniques and reduced genotyping costs make it feasible to score tens of thousands of markers in many individuals (Matsuzaki et al. 2004). The plethora of data has sparked interest in hypothesis testing for identifying genetic markers associated with the phenotype (e. et al. 2005). Testing for marker-phenotype associations is done within the context of a specific experimental design. The experimental design controls the structure of the population under consideration and is therefore a critical component to account for in subsequent modeling and testing. When the genetic structure of the population is under experimental control testing marker-phenotype association is often referred to as QTL mapping (Doerge 2002). In QTL mapping, the issue of which test statistics to use to detect main effects has been discussed …