Student Seminar by Lu Cheng

Please Note: This seminar will be given online.

First-Order Correction of Statistical Significance for Screening Two-Way Epistatic Interactions

We study computational approaches for detecting epistasis (SNP-SNP interactions) that are characterized by genetic disease models (DM) of two or multiple levels of risks. The procedures generally include two steps, first, to find a proper DM for each SNP-pair; second, to test all pairs against the disease outcome. Existing methods such as multifactor dimensionality reduction (MDR) and Wan’s “ratio split” (RS) method relies on the data in both steps, the nominal statistical evidence of which is thus inflated. Corrections are available but computationally expensive for genome-wide studies, so we provide a first-order correction that can be applied in practice with essentially no additional computational cost. We show through simulation studies that such a correction is quite effective in improving the SNP-pair detection for the existing methods. When the correction is applied to our sequential ratio split or merge procedures that are designed for multi-level epistasis detection, the improvement is also noticeably evident, and more evident than that from an approximate application of the recent post-model selection test.