Computational Data-Mining Methods Applied to Combinatorial Chemistry Libraries
Chemoinformatics is the representation and manipulation of chemical information that describes molecular fragments, molecules, and compound libraries. The correlation of chemoinformatic data to biological response data-such as activity toward a target receptor, molecular or cellular toxicity, and the pharmaceutically relevant processes of absorption, distribution, metabolism, and excretion 0-8493-0815-l/04/$0.00+$1.50
(collectively called ADME properties)—can be achieved via data-mining techniques. Data-mining techniques assume that a pattern exists between a population of molecules, their molecular properties, and their biological behavior. Data-mining algorithms use different strategies to uncover or “mine” these patterns. High-throughput methods geared toward pharmaceutical application and therapeutic-target research, such as combinatorial chemistry (CC) and high-throughput screening (HTS), produce large populations of molecules. The associated data overwhelm traditional quantitative structure activity relationship (QSAR) and computational modeling techniques. Thus, high-throughput computational modeling techniques, such as data mining, have become a necessary and natural complement to the present high-throughput combinatorial age. Prioritization of tractable chemical libraries from a large virtual chemical space of possible synthesizable compounds is a necessary and key step in using high-throughput methods to search for biological activity of a given receptor target. Prioritization of biological screening of existing-compound libraries can be based on computational model scores of similarity to known active compounds or predictions of activity, selectivity, and other relevant properties. Strategic application of data-mining techniques aims to focus high-throughput synthesis and screening resources efficiently on biologically relevant or enriched libraries at the “virtual” stage, thus enabling a higher return on efforts.