The Most Informative Spacing Test Effectively Discovers Biologically Relevant Outliers or Multiple Modes in Expression

• Activity Number: 306
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Iwona Pawlikowska*+ and Gang Wu and Michael Edmonson and Zhifa Liu and Tanja Gruber and Jinghui Zhang and Stanley Pounds
• Companies: St. Jude Children's Research Hospital and St. Jude Children's Research Hospital and St. Jude Children's Research Hospital and St. Jude Children's Research Hospital and St. Jude Children's Research Hospital and St. Jude Children's Research Hospital and St. Jude Children's Research Hospital
• Keywords: variable selection for cluster analysis ; outlier detection ; bimodality ; RNA-seq ; cancer genomics

Abstract

Several outlier and subgroup identification statistics (OASIS) have been proposed to discover transcriptomic features with outliers or multiple modes in expression that are indicative of distinct biological processes. Here, we borrow ideas from the OASIS methods in the bioinformatics and statistics literatures to develop the most informative spacing test (MIST) for unsupervised detection of such transcriptomic features. For each individual expression variable, MIST computes the differences between consecutive order statistics (spacings) and multiplies each spacing by the geometric mean of the sizes of the two groups it defines. The spacing with the largest value of this statistic is considered to be the most informative spacing and its significance is determined by simulation. In a pediatric leukemia study, MIST more effectively identified features that divide patients according to gender or the presence of a prognostic fusion-gene in both RNA-seq and microarray expression data than any other OASIS method. MIST may be generalized to identify transcriptomic features that divide subjects into more than two groups and to select features for class discovery analysis.