Two-Stage Designs for Identifying Pathogenic Species in Microbiomic Studies

• Activity Number: 456
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Xiaoshan Wang*+ and Jacqueline Starr and Frias-Lopez Jorge
• Companies: Forsyth Institute and Forsyth Institute and Forsyth Institute
• Keywords: integrated two-stage design ; large scale hypothesis testing ; false discovery rate ; hidden Markov model

Abstract

In microbiomic studies to identify pathogenic bacterial species, an integrated two-stage design, in which a metagenomic technique is used in the first stage, and another technique (e.g. qPCR) in the second stage, may be more cost effective. However, such a design challenges available statistical methods for controlling false discovery rate, since integration requiring to pool data from the two different lab techniques is impractical. Possible strong relationships among microbes also necessitate better statistical techniques for a large scale hypothesis testing. With an assumption on interspecies dependence related to phylogentic similarity, we investigated a procedure of combining local indexes of significance (LISs) in the HMM-FDR framework. We propose a simple average of LISs from the two stages. It can be shown that the average of LIS acts as a pseudo LIS. Then the usual method on LIS for finding significant tests can follow. Simulations indicate our method provide well controlled FDR even if there are strong correlations among tests.