Title: Independence hypothesis weighting in biostatistical practice
Authors: Dominic Edelmann - DKFZ Heidelberg (Germany) [presenting]
Axel Benner - Deutsches Krebsforschungszentrum Heidelberg (Germany)
Abstract: Testing in molecular data often involves a very a large number of hypotheses. In these scenarios, controlling the false discovery rate via Benjamini-Hochberg adjustment of the p-value can lead to very few rejections. To increase the number of discoveries, practitioners commonly apply an unsupervised filtering procedure concentrating the analysis on the most promising hypotheses. However, this filtering procedure is often used quite arbitrarily. Recently, a data-driven technique has been developed, which automatically selects weights for multiple hypothesis testing using side information. Optimizing a target criterion based on the number of rejections, these techniques can dramatically increase the number of discoveries. Beside its complexity compared to unsupervised filtering, one of the main reasons for the disregard of these approaches is that many researchers are not aware that they have important side information available. We present the Independent Hypothesis Weighting (IHW) method and other recently developed data-driven techniques. Moreover, we demonstrate that useful side information is nearly always available for molecular data. Potential covariates include scale and location parameters, gene annotation and text-mining data. We show that this information considerably increases the number of discoveries in large-scale genetic applications.