LAUSR

Molecular dynamics simulations of proteins depend critically on the underlying force field, which may be parameterized against experimental data or high-quality quantum calculations. Here, we develop search algorithms, based on Monte Carlo and steepest descent calculations to optimize the backbone dihedral angle parameters from a single reference simulation. We apply these tools to improve the agreement between simulations of single, capped amino acids and experimentaly determined J-values and secondary structure propensities of these molecules. The parameters are further refined based on simulations of a set of seven proteins and finally validated in simulations on a large set of 52 protein structures. Improvements in the dihedral angle distributions are observed and structural propensities of the proteins are reproduced very well. Overall, the GROMOS 54A8_bb parameter set forms a improvement to previous parameter sets, both for small molecules and for protein simulations.