Performance of the AMBER94, MMFF94, and OPLS-AA Force Fields for Modeling Organic Liquids

Current "second-generation" force fields have been developed for use in Monte Carlo and molecular dynamics simulations for organic and biomolecular systems in solution. The performance of three such force fields, AMBER94, MMFF94, and OPLS-AA, is tested here through Monte Carlo (MC) statistical mechanics simulations for liquid butane, methanol, and N-methylacetamide (NMA). These liquids were chosen for their range of polarity and their representation of typical interactions found in proteins. From simulations at constant temperature and 1 atm pressure, the average errors in the computed densities and heats of vaporization are 3.8% and 5.1% with AMBER94 and 1.5% and 1.3% with OPLS-AA. With MMFF94, the computed densities and heats of vaporization for methanol and NMA are too low by 21−27%, while butane at −0.5 °C does not stop expanding during the simulation. A MC simulation using MMFF94 for butane with the density fixed at the experimental value yields a 32% error in the energy of vaporization. Analyses show that the optimal C−H and H−H nonbonded interactions with MMFF94 are too weak and occur at too large separations. Scaling of the MMFF94 parameter Ai, which is linearly related to the atomic radii, by 0.92 yields improved results for liquid butane.