Accurate determination of solvation free energies of neutral
Posted on：
2022-02-23 19:06
The primary goal of molecular simulations is to accurately predict experimental observations of molecular systems. In the long term, one of their goals is also to develop computational models that are applicable to arbitrary neutral organic molecules and are virtually independent of experimental data. The most important parameter for biophysical synthesis is the free energy, and it is also the most difficult one to predict.
Modern, areal molecular force fields in simulation packages that allow free energy calculations derive some or all of their parameters by fitting them to empirical observations, but with two drawbacks. First, available experimental data are insufficient to produce models that accurately describe existing compounds, and second, error correction of empirical models is difficult because it is hard to identify the source of the error. In contrast, Quantum Mechanical (QM)-parametrized physics-based molecular models can overcome these two issues. Quantum mechanical calculations do not depend on experimental data and could be obtained for any arbitrary moleculesand, and prediction errors can be traced to the imprecise description of the interaction energies and rectified in the model.
In this work, Professor Michael Levitt and co-authors developed a polarization force field model, ARROW FF to predict the free energy of solvation of various neutral organic compounds based on quantum mechanical calculations of the first-nature principle. They implemented a QM-parametrized force field in a simulation stack that covers arbitrary organic molecules and predicts solvation free energies of molecular systems to accuracy of ~0.3 kcal/mol for neutral species, and they demonstrate the predictive ability of the model and simulation machinery by computing solvation free energies for a wide range of chemical functional groups in water and cyclohexane.
The ARROW FF model is multipolar and polarizable, and its computational efficiency is comparable to the most advanced software of its kind currently available. It is notable that nuclear quantum effects (NQE) plays an important role in the free energy calculation and proper accounting of the quantum nature of nuclear motion systematically shifts the predictions toward the experimental values and improves the prediction error of MAE from 0.78 (yellow dot) to 0.2 (blue) kcal/mol. To compare with two widely used empirical models GAFF and AMOEBA (the former is a representative of many fixed-charge models, the latter is a polarizable model), the prediction accuracy of ARROW FF is more than three times that of the two.
This work represents a major advance in biomolecular simulation by constructing a wide-coverage molecular modelling toolset from first principles with excellent predictive ability in the liquid phase.
