Sparse Semisupervised Methods for Predicting Patient Survival Probabilities with Large-Scale Biological Data

• Activity Number: 479
• Type: Contributed
• Date/Time: Wednesday, August 4, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Karthik Devarajan*+
• Companies: Fox Chase Cancer Center
• Keywords: gene expression microarrays ; censored survival data ; semi-supervised learning ; nonnegative matrix factorization ; principal components ; sparse

Abstract

Current efforts in cancer research focus on predicting responses of patients by analyzing gene expression patterns and identifying genes that cause specific responses, with a view to personalizing treatment. In the past decade, microarray technology has made it possible to simultaneously measure the expression levels of tens of thousands of genes. Recently, there has been an increased interest in linking such large-scale biological data with censored survival times to predict the survival probability of a patient. In this paper, we propose semi-supervised methods that combine learning theoretic methods with survival models for censored data. Specifically, we investigate the applicability of sparse variants of nonnegative matrix factorization and principal component analysis in this context. We illustrate our methods via real-life cancer microarray data as well as simulations.