Accurate determination of solvation free energies of neutral organic compounds from first principles

Abstract

The main goal of molecular simulation is to accurately predict experimental observables of molecular systems. Another long-standing goal is to devise models for arbitrary neutral organic molecules with little or no reliance on experimental data. While separately these goals have been met to various degrees, for an arbitrary system of molecules they have not been achieved simultaneously. For biophysical ensembles that exist at room temperature and pressure, and where the entropic contributions are on par with interaction strengths, it is the free energies that are both most important and most difficult to predict. We compute the free energies of solvation for a diverse set of neutral organic compounds using a polarizable force field fitted entirely to ab initio calculations. The mean absolute errors (MAE) of hydration, cyclohexane solvation, and corresponding partition coefficients are 0.2 kcal/mol, 0.3 kcal/mol and 0.22 log units, i.e. within chemical accuracy. The model (ARROW FF) is multipolar, polarizable, and its accompanying simulation stack includes nuclear quantum effects (NQE). The simulation tools’ computational efficiency is on a par with current state-of-the-art packages. The construction of a wide-coverage molecular modelling toolset from first principles, together with its excellent predictive ability in the liquid phase is a major advance in biomolecular simulation.

Explainable Solvation Free Energy Prediction Combining Graph Neural Networks with Chemical Intuition

“…GNN predictions vs experimental values for (a) water and (b) cyclohexane solvation free energies on a data set of small organic molecules from ref . GNN models were trained on solvent holdout data sets with 10% of available water or cyclohexane solvation free energy data added (see the main text).…”

“…GNN models were trained on solvent holdout data sets with 10% of available water or cyclohexane solvation free energy data added (see the main text). ARROW-FF calculated solvation free energies are shown as crosses. The gray shaded area indicates ±0.5 kcal mol –1 within the experiment.…”

“…The solvation Gibbs free energy (Δ G solv ) of a molecule is an essential physicochemical property which governs its behavior in solution. Accurate prediction of a molecule’s Δ G solv value has far-reaching applications in fields ranging from organic synthesis and battery technologies to biological processes. − Molecular simulations for predicting solvation Gibbs free energy via molecular dynamics (MD) or quantum mechanical (QM) calculations, though accurate, can be a time-consuming process due to the need to parameterize new molecules and adequately sample competing solute–solvent interactions and/or entropic effects. − There is much work being conducted in this area to improve the efficiency of these calculations. , Nonetheless, as the chemical space of promising solvent–solute combinations increases with a growing number of desired applications, more computationally efficient approaches such as quantitative–structure activity relationship (QSAR) models and machine learning (ML) have been investigated for rapid screening of databases. , …”

Ab initio machine learning of phase space averages

“…1. Solvation model based on density 52 (SMD) at M06-2X 78 /Def2-TZVPP 67-70 (timing for SMD implemented in Gaussian 79 not to be published) To complete the picture, we also included literature values for FreeSolv concerning the methods ARROW-PIMD8 26 , Thermodynamic integration (TI) with GAFF2 5,6 extracted from the FreeSolv 46 database and reference interaction site model 58,90-92 (3D-RISM).…”

“…Ab initio molecular dynamics (AIMD) simulations not only allow studying molecules but also chemical reactions 2,[23][24][25] . However, they are much more costly than force fields 1,5,6,26,27 due to having to solve approximate quantum mechanical equations at every time step. To this account hybrid set-ups E MD −−→ R ML − − → A using both atomistic simulation and machine learning (ML) have been introduced uniting quantum mechanical equations with surrogate learning on the fly potentials [28][29][30] .…”