Using recursive partitioning analysis to evaluate compound selection methods.

S. Stanley Young, Douglas M. Hawkins

Research output: Contribution to journalArticle

18 Scopus citations

Abstract

The design and analysis of a screening set for high throughput screening is complex. We examine three statistical strategies for compound selection, random, clustering, and space-filling. We examine two types of chemical descriptors, BCUTs and principal components of Dragon Constitutional descriptors. Based on the predictive power of multiple tree recursive partitioning, we reached the following tentative conclusions. Random designs appear to be as good as clustering and space-filling designs. For analysis, BCUTs appear to be better than principal components scores based upon Constitutional Descriptors. We confirm previous results that model-based selection of compounds can lead to improved screening hit rates.

Original languageEnglish (US)
Pages (from-to)317-334
Number of pages18
JournalMethods in molecular biology (Clifton, N.J.)
Volume275
DOIs
StatePublished - 2004

Cite this