Model-Based Clustering with Flipping to Detect and Correct Switched Gene Expression Measurements

• Activity Number: 72
• Type: Contributed
• Date/Time: Sunday, August 9, 2015 : 4:00 PM to 5:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): William Young* and Adrian Raftery and Ka Yee Yeung
• Companies: and University of Washington and University of Washington
• Keywords: Genomics ; Bayesian Statistics ; Clustering ; Gene Expression

Abstract

We investigate clustering methods applied to a large gene expression dataset provided by the NIH Library of Integrated Cellular Signatures (LINCS). These data consist of 1.4 million experiments, each measuring the gene expression levels of a cell line with a particular perturbation applied. In this talk, we will address some of the statistical questions from this data. In particular, we consider the situation where two genes are measured together using the same analyte and then deconvolved, assigning expression levels to each gene separately. We observed that this process sometimes results in the expression levels of the two genes being switched and so assigned to the wrong gene. We describe a model-based clustering method involving flipping of clusters to detect and correct these errors. This method results in an improved, more consistent dataset.