Analysis of a Large Structure/Biological Activity Data Set Using Recursive Partitioning

Abstract

Combinatorial chemistry and high-throughput screening are revolutionizing the process of lead discovery in the pharmaceutical industry. Large numbers of structures and vast quantities of biological assay data are quickly being accumulated, overwhelming traditional structure/activity relationship (SAR) analysis technologies. Recursive partitioning is a method for statistically determining rules that classify objects into similar categories or, in this case, structures into groups of molecules with similar potencies. SCAM is a computer program implemented to make extremely efficient use of this methodology. Depending on the size of the data set, rules explaining biological data can be determined interactively. An example data set of 1650 monoamine oxidase inhibitors exemplifies the method, yielding substructural rules and leading to general classifications of these inhibitors. The method scales linearly with the number of descriptors, so hundreds of thousands of structures can be analyzed utilizing thousands to millions of molecular descriptors. There are currently no methods to deal with statistical analysis problems of this size. An important aspect of this analysis is the ability to deal with mixtures, i.e., identify SAR rules for classes of compounds in the same data set that might be binding in different ways. Most current quantitative structure/activity relationship methods require that the compounds follow a single mechanism. Advantages and limitations of this methodology are presented.

Chemogenomics: an emerging strategy for rapid target and drug discovery

“…Statistical and chemoinformatics approaches for assessing the quality control of HTS data and for mining their chemical and biological information have been developed, for example, by incorporating pattern-detection methods for the identification of pipetting artefacts or for the detection of chemical-class-related effects. The development of chemoinformatics methods and procedures, such as RECURSIVE PARTITIONING, PHYLOGENETIC-LIKE TREE ALGORITHMS or BINARY QUANTITATIVE STRUCTURE-ACTIVITY RELATIONSHIPS (QSARS) [81][82][83][84][85][86][87] , which support the automatic identification of hits that are frequently identified by HTS, false positives and negatives, as well as structure-activity relationship (SAR) information, is essential for generating knowledge from HTS data 83,[88][89][90] . The myriad efforts that surround the design of appropriate analytic tools have to cope with the difficulty of integrating disparate types of information, especially PARSING and assimilating both chemical and genomics information data (tools such as Scitegic Pipeline pilot or Kensington Inforsense provide the required integration concepts (see online links box)).…”

Assessing Different Classification Methods for Virtual Screening

“…Decision trees, therefore, represent Boolean functions, even in the case of a larger range of outputs. 34 The version used here is Quinlan's C4.5. 35 A decision tree is an arrangement of tests that prescribes an appropriate test at every step in an analysis.…”

Classification of Kinase Inhibitors Using a Bayesian Model

“…2325 Stanley Young and colleagues implemented a recursive partitioning method a few years ago that can be used to derive predictive models to distinguish active from inactive compounds. 20,21 This method, compared to some older implementations, has the advantage of being able to handle a large number of descriptors and very large compound data sets.…”