Truncated and Weighted Transformation Methods for Combining P-Values

• Activity Number: 642
• Type: Contributed
• Date/Time: Thursday, August 8, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Zheyang Wu*+ and Tiejun Tong
• Companies: WPI and Hong Kong Baptist University
• Keywords: hypothesis testing ; combination p-values ; truncation ; GWAS

Abstract

In genetics and epigenetics it is critical to develop powerful statistical methods to identify associations between a response and sets of genetic markers or genes. From statistical perspective, due to the threshold effect on the signal strength, when comparing with the identification of individual signals, the identification of some signals existing in a group of candidates has weaker requirements for signal strength and density. For combining individual p-values, proper procedures of truncating, weighting, and transforming the p-values can potentially help signals standing out, and thus improve statistical power. In this research, we give a unified study of p-value combination methods by defining a general transformation function that covers various truncation and weighting procedures. Theoretical results and simulation show that the quadratic function of truncated P-value transformation provide a higher statistical power than other methods in most setups of signal strength and density. The methods are applied and compared in real data analysis of genome-wide association studies.