The detection of rare deleterious variants is the pre-eminent current technical challenge in statistical genetics. In previous work we have used hierarchical modeling techniques to estimate the relative risks of individual rare variants from a known risk gene. Since each specific variant is of crucial interest to the individuals and their family members who possess this specific variant, classifying each of these variants as harmful versus harmless is a particularly important but challenging goal because of the sparseness of the evidence for each individual variant. Using simulations we examine the properties of different false discovery controlling procedures in this setting with the goal of optimizing the classification of rare variants as deleterious versus neutral. We illustrate the methods with an application to a real study of breast cancer.
This is joint work with Venkatraman Seshan and Colin Begg.