GLAD: A Mixed-Membership Model for Heterogeneous Tumor Subtype Classification

• 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): Patrick Flaherty* and Hachem Saddiki and Jon McAuliffe
• Companies: University of Massachusetts Amherst and Worcester Polytechnic Institute and UC Berkeley
• Keywords: mixed membership model ; intratumor heterogeneity ; cancer ; genomics ; variational inference

Abstract

Genomic analyses of many solid cancers have demonstrated extensive genetic heterogeneity between as well as within individual tumors. However, statistical methods for classifying tumors by subtype based on genomic biomarkers generally entail an all-or-none decision, which may be misleading for clinical samples containing a mixture of subtypes and/or normal cell contamination. We have developed a mixed-membership classification model, called GLAD, that simultaneously learns a sparse biomarker signature for each subtype as well as a distribution over subtypes for each sample. We demonstrate the accuracy of this model on simulated data, in-vitro mixture experiments, and clinical samples from the Cancer Genome Atlas (TCGA) project. We show that many TCGA samples are likely a mixture of multiple subtypes.