A Comparison of Node-Splitting Rules in Recursive Partitioning Analysis of Multivariate Quantitative Structure Activity Data

• Activity Number: 377
• Type: Contributed
• Date/Time: Wednesday, August 1, 2007 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Yujun Wu*+ and Marc Genton and Leonard A. Stefanski
• Companies: University of Medicine and Dentistry of New Jersey and University of Geneva and North Carolina State University
• Keywords: QSAR ; classification tree ; Hotelling's T2 ; MultiSCAM ; pooled component test

Abstract

Tree-structured methods have been advocated to model non-linear relationships in quantitative structure-activity relationship (QSAR) studies. One such algorithm, MultiSCAM (Keefer, 2001), was developed to analyze QSAR data with multivariate continuous responses, classifying objects into similar categories by building a tree with recursive partitioning. Hotelling's T2 test is the method used to determine splits in MultiSCAM. However, this test is not feasible when the dimension of an observation exceeds the number of observations. This problem is further exacerbated by the fact that missing values are common. We consider two alternatives, the pooled component test (PCT) proposed by Wu et al. (2006), and a simple ANOVA F test. To compare the three node-splitting tests, we introduce a comprehensive simulation design. A drug discovery data set is used to illustrate the methods.