Smoothed Stability Selection for Analysis of Sequencing Data

• Activity Number: 613
• Type: Contributed
• Date/Time: Thursday, August 2, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Eugene Urrutia*+ and Li Yun and Michael C Wu
• Companies: The University of North Carolina at Chapel Hill and The University of North Carolina at Chapel Hill and The University of North Carolina at Chapel Hill
• Keywords: variable selection ; smooth stability selection ; next generation sequencing ; rare variants ; high dimensional data ; resampling methods

Abstract

High dimensional data are increasingly common. Difficulties in model interpretation and limited power to detect effects have led to the use of variable selection methods, including penalized regression methods such as the LASSO. Recent advances in the variable selection literature suggest that resampling strategies which include stability selection, complementary stability selection, and bolasso, can offer improvements over the LASSO and non-resampling based approaches. We show that common resampling based methods can be recast as LASSO with reweighted observations where weights are discrete and identically distributed from a specified distribution. Sequencing data presents an additional challenge in that many predictors are rare (minor alleles observed in only a few individuals) and have a high probability of exclusion in the resampling schemes. Thus, we have developed the smooth stability selection procedure where we replace the discrete weights with continuous weights, and thus avoid excluding rare predictors. Simulation results and real data analyses suggest that our proposed method increases power to select rare variables while retaining type I error control.