Prediction of Charged Small Molecule Conformations in Solution using a Balanced ML/MM Potential

Reliably evaluating the most stable conformations of charged small organic molecules in solution poses a major challenge to computational chemistry, due to the accuracy or computational efficiency limitations of conventional methodologies. In this paper, we present a hybrid machine learning/molecular mechanics (ML/MM) potential, with dynamic, conformationally-dependent charges on the atoms in the ML region, that can be used to address this important challenge in drug design. By way of a case study, molecular dynamics simulations were used to compare the performance of intermolecular potentials in evaluating the solution phase conformational free energy differences of pharmaceutically relevant ligands based on the protonated 2-phenylethylamine scaffold. We show that satisfactory performance of an MM-only force field in predicting the preferred conformations of these molecules in solution (with a mean absolute error (MAE) = 0.9 kcal/mol) is due a convenient cancellation of errors in the intramolecular and intermolecular contributions to the conformational free energy difference. Specifically, the underestimation of intramolecular cation- interactions which preferentially stabilise the folded conformation is compensated by underestimated solute-solvent interactions which preferentially stabilise the extended conformation. A straightforward approach to an ML/MM potential, in which the intramolecular MM representation of the solute is simply substituted by a ML model without making other changes to the potential, disrupts this error cancellation and results in a systematic underperformance relative to the MM-only approach, with a MAE of 1.9 kcal/mol. To remedy this shortcoming, a new ML/MM potential is presented that combines an accurate, trained-for-purpose ML (PairFEQ-Net) model of the charged solute, with the polarisable SWM3 MM model of water. Crucially, by employing an empirical parameter to scale gas phase charges, the ML model predicts dynamic, solvent-polarised charges that effectively electrostatically embed the solute in the solvent in a balanced way, outperforming the MM-only approach when it comes to conformational free energy predictions, reducing the MAE to 0.5 kcal/mol. This balanced ML/MM potential furnishes a possible route to the chemically accurate prediction of the shapes of organic molecules in aqueous solution.