Critical assessment of docking programs and scoring functions


Gregory L. Warren1, Webb Andrews III2, Anna Maria Capelli2, Brian P. Clarke2, Judith M. LaLonde2, Millard H. Lambert2, Mika Lindvall2, Neysa Nevins2, Catherine E. Peishoff1, Simon F. Semus2, Stefan Senger2, Giovanna Tedesco2, Ian D Wall2, James M. Woolven2, and Martha S. Head1. (1) Computational, Analytical and Structural Sciences, GlaxoSmithKline Pharmaceuticals, Research and Development Division, P.O. Box 5089, UP1110, Collegeville, PA 19426-0898, (2) Computational, Analytical and Structural Sciences, GlaxoSmithKline, 5 Moore Drive, P.O. Box 13398, Research Triangle Park, NC 27709
With the recent dramatic increase in available structural data has come a need to evaluate the current state of the art for docking and scoring algorithms. In an effort to understand the strengths and weaknesses of such algorithms, we have undertaken an evaluation of the performance of 10 docking programs and 37 scoring functions against 8 proteins from 7 evolutionarily diverse targets for three tasks: accuracy of binding mode prediction, data enrichment during virtual screening, and the ability to rank order by affinity for lead optimization. While performance for any particular docking program varied across the targets evaluated, docking programs and scoring functions are able to reproduce crystallographically observed binding modes and identify active compounds from a pool of decoy compounds. However, current docking programs and scoring functions are unable to rank order compounds by affinity.

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