Model-Based Classifications of High-Throughput Data Review, Design, and Application to a Cancer Clinical Study

• Activity Number: 377
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): AC Cambon*+ and Shesh Nath Rai
• Companies: Department of Bioinformatics and Biostatistics, University of Louisville and University of Louisville
• Keywords: classification ; dimension reduction ; sample size planning ; high thoughput data ; predictive modeling ; machine learning

Abstract

It is now widely recognized that many treatments for cancers such as melanoma are effective for only a subset of a population. However clinical studies for cancer treatments are more often powered to detect an overall treatment effect. In this study, parametric high throughput classification methods which could be used in a clinical study to identify a subset of patients more sensitive to treatment or more prone to relapse are reviewed. Methods such as logistic classification, linear discriminant analysis, and ROC regression, along with appropriate dimension reduction methods, are discussed. Approaches to sample size planning which take into account high throughput data and dimension reduction are also reviewed.