Improving the enrichment of high-throughput docking results using machine learning


Anthony E. Klon, Meir Glick, and John W. Davies. Lead Discovery Center, Novartis Institutes for Biomedical Research, 100 Technology Square, Cambridge, MA 02142
High-throughput docking (HTD) is a commonly utilized technique in the drug discovery process. However, the ability to accurately rank compounds using a scoring function remains problematic. Here we show that by employing a simple machine learning method (na´ve Bayes), it is possible to significantly improve the ranking of compounds and thereby the accuracy of HTD. Three protein targets were reviewed using three software packages; Dock, FlexX and Glide. For each target, known active compounds and the Available Chemical Database (ACD) were evaluated. In cases where HTD alone was able to produce enrichment of known actives, the application na´ve Bayes was able to significantly improve upon the enrichment. The application of the na´ve Bayes classifier to enrich HTD results can be carried out without any a priori knowledge of the active compounds. The methodology results in superior enrichment of known actives compared to the use of HTD and consensus scoring alone.

Docking and Scoring
9:00 AM-12:20 PM, Tuesday, August 24, 2004 Pennsylvania Convention Center -- 109B, Oral

Division of Computers in Chemistry

The 228th ACS National Meeting, in Philadelphia, PA, August 22-26, 2004