Understanding Force-Field Bias in Pin1WW

Abstract

Force field-based molecular dynamics (MD) is a powerful tool that has produced many key insights into the mechanisms of protein folding. However, MD is not without its flaws. One well-documented problem with MD is that some force fields have unphysical biases towards certain secondary structures. This project focuses on understanding one particular example of force field bias: folding pin1WW with the CHARMM27 force field. Pin1WW is a fast-folding protein made up of three β-sheets connected with two hairpin turns. Previous, state-of-the-art folding simulations of pin1WW using the CHARMM27 force field only resulted in α-helical structures, a clear indication of force field bias. This work focuses on two complementary aspects of force field bias: 1) detecting it before performing a simulation and 2) alleviating it to perform unbiased simulations. The initial attempts to detect the presence of bias were based on typical scoring functions used to determine low energy geometries of protein structures. These scoring functions were applied to native state as well as several nonnative helical structures. Many typical functions were not sensitive enough to detect force-field bias. Attempts to correct force-field bias will be based on density functional theory (DFT). Namely, the force matching algorithm will be used to change the force field so that it reproduces DFT energies and forces. However, density functionals need to be benchmarked to ensure that they give accurate results for proteins. Therefore, we also present a comparison of BLYP and PBE (with and without empirical dispersion corrections) and M06.