Analysis of a Large StructureBiological Activity Data Set Using Recursive Partitioning and.pdf

Analysis of a Large Structure/Biological Activity Data Set Using Recursive Partitioning and Simulated Annealing Student: Ke Zhang MBMA Committee: Dr. Charles E. Smith (Chair) Dr. Jacqueline M. Hughes-Oliver Dr. Jason A. Osborne Advisor for this project: Dr. Jacqueline M. Hughes-Oliver 02/2003 Project Report for MBMA - Institute of Statistics Mimeo Series #2542 Analysis of a Large Structure/Biological Activity Data Set Using Recursive Partitioning and Simulated Annealing ABSTRACT Large quantities of structure and biological activity data are quickly accumulated with the development of high-throughput screening (HTS) and combinatorial chemistry. Analysis of structure-activity relationships (SAR) from such large data sets is becoming challenging. Recursive partitioning (RP) is a statistical method that can identify SAR rules for classes of compounds that are acting through different mechanisms in the same data set. We use a newly proposed method called Recursive Partitioning and Simulated Annealing (RP/SA) to produce stochastic regression trees for biological activity. In the new algorithm a set of structural descriptors is extracted at each splitting node by using SA as a stochastic optimization tool. For one data set, results show that RP/SA is advantageous in analyzing the SAR information. INTRODUCTION With the development of high throughput screening (HTS) techniques1,2, large structure/biological activity data sets are becoming available. Consequently, effective methods that can manage this amount of data and convert it into utilizable information are in great