Non-Specific Filtering of Beta-Distributed Data

• Activity Number: 614
• Type: Contributed
• Date/Time: Thursday, August 7, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Xinhui Wang*+ and Peter W. Laird and Toshinori Hinoue and Susan Groshen and Kim Siegmund
• Companies: and USC Keck School of Medicine and University of Southern California and University of Southern California and University of Southern California Keck School of Medicine
• Keywords: feature selection ; DNA Methylation ; cluster analysis ; Beta distributed data ; variance-stabilizing transformation

Abstract

Non-specific feature selection is a dimension reduction procedure performed prior to cluster analysis of high dimensional data. DNA methylation is measured as a proportion, bounded between 0 and 1, with variance a function of the mean. Filtering on standard deviation biases the selection to probes with mean values near 0.5. We developed and compared alternate filter methods using simulation as well as real Illumina's Infinium HumanMethylation data. We found that for data having a small fraction of samples with abnormal methylation of a subset of normally unmethylated CpGs, a novel filter statistic that utilized a variance-stabilizing transformation for Beta distributed data outperformed the standard deviation filter of beta value, or its log-transformed M-value, in detecting cancer subtype in a cluster analysis. Despite mainly different features enriched, standard deviation filter in conjunction with cluster analysis did result in sample subsets that overlapped those found by our novel filter. Since cluster analysis is for discovery, we would suggest trying both filters on new data sets, evaluating the overlap of features selected and clusters discovered.