Permutation-Based Maximum Covariance Analysis (PMCA)

• Activity Number: 607
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Robyn L. Ball* and Yasuhiro Fujiwara and Fengyun Sun and Jianjun Hu and Mary Ann Handel and Gregory W. Carter
• Companies: The Jackson Laboratory and The Jackson Laboratory and The Jackson Laboratory and The Jackson Laboratory and The Jackson Laboratory and The Jackson Laboratory
• Keywords: Nonparametric ; High dimensional data ; Maximum Covariance Analysis ; Paired Data

Abstract

Maximum Covariance Analysis (MCA) is often an effective tool for detecting common signals in two sets of variables. However, MCA can be limited by a tendency to fit spurious patterns when faced with increased sampling variation. To minimize fitting spurious correlations, current methods employ a parametric smoothing procedure that requires a normality assumption. We introduce a novel permutation-based extension of maximum covariance analysis (PMCA) that not only overcomes the spurious pattern identification liability but that does so without the need for any parametric assumptions. An estimated false positive rate (FPR) is specified and through an iterative permutation procedure, the algorithm chooses an optimal window and maps one dataset onto a second dataset obtained from the same samples. The primary motivation for developing PMCA was to map gene expression data to specific cell stages determined by quantitative cytological analyses, and thereby define the meiotic transcriptome. We have also demonstrated the utility of PMCA for other types of paired data.