A critical evaluation of several global optimization algorithms for the purpose of molecular docking

Several global optimization algorithms were applied to the problem of molecular docking: random walk and Metropolis Monte Carlo Simulated Annealing as references, and Stochastic Approximation with Smoothing (SAS), and Terminal Repeller Unconstrained Subenergy Tunneling (TRUST) as new methodologies. Of particular interest is whether any of these algorithms could be used to dock a database of typical small molecules in a reasonable amount of time. To address this question, each algorithm was used to dock four small molecules presenting a wide range of sizes, degrees of flexibility, and types of interactions. Of the algorithms tested, only stochastic approximation with smoothing appeared to be sufficiently fast and reliable to be useful for database searches. This algorithm can reliably dock relatively small and fairly rigid molecules in a few seconds, and larger and more flexible molecules in a few minutes. The remaining algorithms tested were able to reliably dock the small and fairly rigid molecules, but showed little or no reliability when docking large or flexible molecules. In addition, to decrease the error in the typical grid-based energy evaluations a new form of interpolation, logarithmic interpolation, is proposed. This interpolation scheme is shown to both quantitatively reduce the numerical error and practically to improve the docking results. c © 1999 John Wiley & Sons, Inc. J Comput Chem 20: 1740–1751, 1999