A SVM-Based Meta-Analytic Approach for Simultaneous Gene Selection and Case Prediction in Alzheimer's Disease

• Activity Number: 647
• Type: Topic Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: International Indian Statistical Association
• Author(s): Samiran Ghosh*+ and Meeta Pradhan
• Companies: Winthrop University Hospital and Indiana University Purdue University Indianapolis
• Keywords: Alzheimer's disease ; Biomarker ; Classification ; Feature selection ; Genomics ; SVM

Abstract

The discovery of biomarkers, as well as classifying subjects in proper disease group is a ubiquitous task. Over the last decade there is an exponential growth in bioinformatics research and publications. When results from several multi-study analysis is combined, the resulting analysis is known as ``Meta-analysis". We propose a very general framework for combining several studies under L_1-norm support vector machine (SVM). L_1-norm SVM has the computational advantage over many other competing classification algorithms. It also inherits the automatic feature selection property owing to its absolute norm penalty. We apply our algorithm for classification and biomarker selection in combining several studies related to Alzheimer's disease (AD). Using the pathway analysis software Pathway Studio, we analyzed the biomarkers with respect to their pathways and their inferences in literature with AD. We are able to identify already established as well as few new pathways and biomarkers of AD. The combined data set preserve most of the top voted biomarkers from separate data sets. Also we reveal some new set of candidate biomarkers only discovered when combining several studies together.