Filtering, Bias, and Biomarker Identification

• Activity Number: 72
• Type: Contributed
• Date/Time: Sunday, August 3, 2014 : 4:00 PM to 5:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Gregory Hather*+ and Ray Liu
• Companies: Takeda and Takeda
• Keywords: filtering ; biomarker ; bias ; false discovery rate

Abstract

Researchers often perform hypothesis testing on a large number of candidate biomarkers. To increase the power of such tests, a filtering step may be applied before the hypothesis testing. For example, with RNA-seq expression analysis, one might exclude the genes which have very few counts across all the samples. Although filtering can reduce the number of hypotheses tested, it can introduce bias in the estimated false discovery rate if the filtering step is not taken into account. Here, we describe several different filtering strategies and determine which ones introduce bias in the false discovery rate. We also consider methods for correcting this bias, as well as alternatives to filtering.