Approaches to the statistical-genetic analysis of association and microarray data

Complex disease results from the combined effects of many genetic and nongenetic factors and should be studied in a global context. This is facilitated by biotechnological advances that allow for genome-wide association studies and microarray expression experiments. There is a critical need to develop appropriate procedures to analyze the large amounts of data generated from these methods. The identification of causative factors is limited by statistical problems that arise from performing multiple tests, the “curse of dimensionality”, and the need to reliably detect small gene effect sizes. These concerns are further complicated by modest sample sizes for association studies and the limited number of replications feasible for microarray experiments. Approaches that address these issues are applied to a prostate cancer microarray experiment and a series of simulations that model the relationship between quantitative trait loci (QTLs) and disease outcome. Results indicate that data-mining techniques can be applied to effectively “learn” the relationships between a large number of possible causative factors and phenotypic outcome, while controlling for type I and II statistical error rates.