A Tree-Based Approach for Gene-Gene Interaction Detection in DNA Sequencing Data

• Activity Number: 246
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Chi Song*+ and Heping Zhang
• Companies: Yale Univeristy and Yale University
• Keywords: Sequencing ; Rare variant ; Tree

Abstract

The Next Generation Sequencing can generate rare variants (frequency< 0.01) in multiple samples, which makes it possible to investigate diseases caused by many rare variants with high relative risks. However, because of the small frequencies of the rare variants, most of the methods are powerless to detect them individually. They are usually combined on gene level to boost the statistical power. We propose a tree-based method to not only select variants to combine, but also detect gene-gene interactions. In our tree model, each splitting node represents a gene. For each node, a subset of variants in the corresponding gene is selected. Based on whether any variant in the subset is observed, a binary decision is made. The subset is selected to optimize the impurity measurement such as entropy. To speed up the computation, the variants can be ordered according to their frequency, local linkage disequilibrium pattern and position. By constructing a tree with multiple genes, the gene-gene interaction is accounted non-parametrically. This tree-based approach can be further developed into a random forest. Both simulation and real data analysis show the advantage of our method.