Force field validation using protein side chain prediction

The prediction of protein side chain conformations is used to evaluate the accuracy of force field parameters. Specifically, new torsional parameters have recently been reported for the OPLS-AA force field, which achieved substantially better accuracy with respect to high level gas-phase quantum chemical calculations [J. Phys. Chem. B 2001, 105, 6474]. Here we demonstrate that these new parameters also lead to qualitatively improved side chain prediction accuracy. The primary emphasis is on the prediction of single side chain conformations, with the rest of the protein held fixed at the native configuration. Errors due to incomplete sampling can thus be essentially eliminated, using a combination of rotamer search and energy minimization. In addition, the protein environment is modeled realistically using implicit solvation and an explicit representation of crystal packing effects. Aided by the development of new algorithms, these calculations have been performed with modest computational requirements (a cluster of PCs) on a database of 36 proteins (similar to5000 total residues). The side chain prediction tests that we employ are quite general and can be used to evaluate nonbonded or solvation parameters as well. As such, they provide a useful complement to decoy studies for force field validation.