Monte Carlo algorithms for docking to proteins

Abstract

The goal of docking is to predict binding interactions between molecules. We are primarily interested in docking as a tool for the structure-based design of new ligands that could serve as lead compounds for drug development. The program BOXSEARCH uses a Monte Carlo algorithm to explore the relative orientation and position of two molecules. Multiple runs are carried out from different random starting positions and orientations, and the temperature of the system is gradually reduced. An unbiased sampling of low energy states is the result. BOXSEARCH has been tested on a number of known complexes, involving both protein and small molecule ligands. Although a better treatment of solvent effects and of flexibility would improve the ranking of results, the complexes can be reconstructed successfully, even using uncomplexed conformations of the molecules. We are currently implementing two major enhancements. First, the code is being rewritten in a more general and adaptable form, using the object-oriented programming language C⁺⁺. Object-oriented programming allows us to reuse code very easily and also lets us use a higher level of abstraction. In practical terms, this makes it much easier to program and test new ideas for molecular simulations, including better treatments of solvent and flexibility. Second, genetic algorithms are being implemented as a more general and powerful optimization tool. We envision simulations in which molecules “evolve” on the computer, by mutation and recombination in the binding site.