The identification of genes that lead to common complex diseases has been fraught with difficulties, including the need for large sample sizes and problems replicating findings. A critical aspect is the large amount of heterogeneity of factors related to common diseases, including multiple genes, environmental risk factors, and possibly their interactions. The need for large sample sizes can increase the amount of heterogeneity, and attempting to replicate findings can be confounded by uncontrolled covariates. Controlling for heterogeneity is crucial to detect the weak to moderate effects of underlying genes. Motivated by a genome-wide scan for susceptibility genes for prostate cancer, we have developed new statistical methods, based on recursive partitioning, in order to partition the linkage data into homogeneous subgroups defined by combinations of covariates. Although recursive partitioning has enjoyed success as a data exploration tool in some areas of scientific inquiry, only a few adaptations have been suggested for its application to genetic linkage analyses. The fundamental theory behind recursive partitioning will be reviewed, in order to illustrate our extensions to multipoint linkage analysis of affected relative pairs, both in terms of statistical methods and computer software implementation. Demonstration of these new methods will be presented by application to an ongoing study of the genetics of prostate cancer conducted at the Mayo Clinic.
Meet the speaker in Room 212 Cockins Hall at 4:30 p.m. Refreshments will be served.