A fast and powerful tree-based association test for detecting complex joint effects in case-control studies

Monday, September 22, 2014 — Poster Session II
4:00 p.m. – 6:00 p.m.	FAES Academic Center	NCI	GEN-18

* FARE Award Winner

Authors

• H. Zhang
• W. Wheeler
• Z. Wang
• P.R. Taylor
• K. Yu

Abstract

Motivation: The SNPs identified through GWAS only explain a small proportion of the heritability for complex diseases. The multivariate test (gene-based or pathway analysis) aggregating the effects among a set of SNPs therefore becomes a promising alternative to the single-marker test that is widely used in current GWAS. The existing multivariate tests derived from the logistic regression model focus on assessing the additive effects of multiple SNPs on a disease outcome and are less sensitive for non-additive joint effect. The tree-structure model is more apt to capture complex effect, but is suffered from computational burden when using the resampling-based procedure to evaluate p-values. Methods: We proposed a TREe-based Association Test (TREAT) that incorporates adaptive model selection to identify the optimal tree representing the joint effect. We designed an ultra-fast algorithm adopting Boolean operation to build the tree model, which speeds up the testing procedure by 60 times. Results: We applied TREAT on over 20,000 genes in a GWAS of esophageal squamous cell carcinoma (ESCC) involving 1942 cases and 2111 controls. TREAT identified a novel association between the gene CDKN2B and ESCC (P = 6.0E-8). We also demonstrated, through simulation studies, the power advantage of TREAT in detecting complex joint effect.

back to top