Detecting Signal Regions in Whole-Genome Association Studies

• Activity Number: 433
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Zilin Li* and Xihong Lin
• Companies: and Harvard School of Public Health
• Keywords: Correlated test statistics ; Multiple hypothesis ; Large deviation ; Scan statistics

Abstract

We consider in this paper detecting signal regions associated with human diseases in whole genome association studies. While common gene- or region-based procedures only test for SNPs in prespecified regions, we propose a chi-squared based scan and segmentation algorithm to detect the existence and location of signal segments. The asymptotic property of the scan statistic allows us to derive an asymptotic threshold to control the familywise error rate. We also show that, under regularity conditions, the proposed procedure consistently selects the true signal regions. Our simulation studies indicate that the proposed procedure has a better finite sample performance over several existing methods, especially in presence of non-signal variants and strong correlation between signal and non-signal variants within signal regions. We apply the proposed procedure to analyze a lung cancer genetic dataset to identify the regions of SNPs which are associated with lung cancer risk.