Dimension Reduction Through Variable Selection: A Fibrosis Case Study

• Activity Number: 238
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Katja Sabine Remlinger*+
• Companies: GlaxoSmithKline
• Keywords: Biomarker ; Cross Validation ; Classification ; RandomForest ; Selection Bias

Abstract

Variable Selection is an important step in prediction modeling of high dimensional data. Not all variables in the data are informative, and removal of irrelevant noise often improves the prediction performance. If data is of biological origin, cost-efficiency and an improved understanding of the underlying biological process that generated the data are additional reasons for performing variable selection. Once important variables have been selected, it is equally important to assess whether these variables can be validate in subsequent studies. Using real data from a Fibrosis study, we will illustrate the benefits and challenges of the initial variable selection and the subsequent validation steps. We used univariate and multivariate analysis techniques to select variables, and assessed their performance by means of predictive power and validation strength in subsequent studies.