In recent years, virtual high-throughput screening (vHTS) techniques have been successfully applied in the drug discovery process. In many cases, these vHTS techniques are leveraged to prioritize subsets of chemical libraries for acquisition and testing in physical screens. However, for computer-aided drug design (CADD), it is advantageous to have algorithms which are capable of designing new chemical entities for a specific biological target. An evolutionary algorithm was implemented as part of the BCL::ChemInfo suite within the Biochemistry Library (BCL), a C++ library developed at Vanderbilt University, to iteratively generate chemical species with high predicted biological activities for use in focused library design for hit-to-lead optimization. Quantitative structure activity relationships based on machine learning techniques were used to predict the biological activity of compounds in each generation. The compounds with the highest predicted activity, as well as a smaller number of lower activity species, were subjected to combination, crossover, and mutation to form the subsequent generation. Termination criteria were based on a percentage of compounds in a single generation achieving a predicted activity above a pre-determined cutoff, or after reaching a pre-set number of generations in the event the first criterion could not be satisfied. This method was benchmarked using a previously published set of 9 datasets designed for the validation of novel CADD methods. Each of the datasets was compiled from publicly available HTS data taken from PubChem, and contain a minimum of 150 active compounds each. In addition, the datasets span a range of protein targets including GPCRs, ion channels, and enzymes. Here we present the results of this focused library design application, BCL::EvoGen.

BCL::EvoGen: An evolutionary algorithm for focused library design

248th National Meeting (2014)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.