Testing statistical association of individual rare variants is underpowered due to low frequency. A common approach is to test the aggregated effects of individual variants in a locus such as genes. Current methods have distinct power profiles that are determined by underlying assumptions about the genetic model and effect size. Here we describe a parametric Bayesian approach to detect the association of rare variants. We express the assumptions about effect size by setting the prior distribution in the model, which can be adjusted based on the experimental design. This flexibility allows our method to achieve optimal power. The algorithmic contribution includes a dynamic program for efficient calculation of the association test statistic. We tested the method in simulated data, and demonstrated that it is better powered to detect rare variant association under various scenarios.
展开▼
