Another consideration is that how genotype is coded implies a biological model, as reviewed above, and, as such, the model should be specified and justified (or explicitly stated as exploratory, with appropriate corrections for multiple testing). Genotypes should not be collapsed purely to increase power and if they are, then effects should be described across all genotype groups for completeness. With increasing recognition that individual genetic polymorphisms on their own are likely to have very small effects, there should be justification provided if single polymorphisms are being studied in isolation. Further, with growing knowledge about gene networks and integrated functional pathways, there is opportunity to think more broadly about the potential for incorporating gene networks or pathways into tests of GxE, allowing one to go beyond focusing on a single gene and instead focus on sets of genes that interact biologically. This brings its own set of complications, as decisions must be made about the nature of genetic effects across the pathway or network: is a mutation in any of the genes in the network sufficient? Are all variations within
