In recent years there has been much focus on the use of single nucleotide polymorphism (SNP) fine genome mapping to identify causative mutations for traits of interest; however, many studies focus only on the marginal effects of markers, ignoring potential gene interactions. Simulation studies have show that this approach may not be powerful enough to detect important loci when gene interactions are present. While several studies have examined potential gene interaction, they tend to focus on a small number of SNP markers. Given the prohibitive computation cost of modeling interactions in studies involving a large number SNP, methods need to be develop that can account for potential gene interactions in a computationally efficient manner. This study adopts a machine learning approach by adapting the ant colony optimization algorithm (ACA), coupled with logistic regression on haplotypes and genotypes, for association studies involving large numbers of SNP markers. The proposed method is compared to haplotype analysis, implemented using a sliding window (SW/H), and single locus genotype association (RG). Each algorithm was evaluated using a binary trait simulated using an epistatic model and HapMap ENCODE genotype data. Results show that the ACA outperformed SW/H and RG under all simulation scenarios, yielding substantial increases in power to detect genomic regions associated with the simulated trait.
genome; simulation; SNP
