Figure 1From: Evolving hard problems: Generating human genetics datasets with a complex etiology Display of the MDR Results for a Three-SNP Interaction. This figure illustrates the solution dataset for a run of our algorithm which attempted to create a three-marker dataset with a high third-order gene-disease association and no lower-level effects. Each square in the plots represents a specific genotypic combination. Within each square the first bar measures the number of cases and the second bar measures the number of controls. The darker squares represent a genotypic combination that was considered high-risk due to the greater number of cases than controls contained within. The top panel, labeled A, shows the relation between each single marker and case-control status. The ability of our algorithm to minimize first-order associations is visible by the relatively equal height of the bars within each square. Of the three one-way associations, X1 versus case-control status scored the highest with an accuracy of 0.502. The middle panel, labeled B, shows the relation between all three two-locus combinations and disease. Again our algorithm succeeded in preventing any major ability to classify disease status based on a specific genotypic combination. The highest two-way effect was between X1, X2 and disease with an accuracy of 0.513. The bottom panel, labeled C, shows the subjects fully decomposed into all genotypic combinations illustrating the third-order effect. Under this level of analysis, each genotypic combination expresses great ability to differentiate between cases an controls. As desired, the accuracy was high at 0.804.Back to article page