Computing solvation free energies of small molecules with first principles accuracy

Abstract: Free energies play a central role in characterising the behaviour of chemical systems and are among the most important quantities that can be calculated by molecular dynamics simulations. Solvation free energies in various organic solvents, in particular, are well-studied physicochemical properties of drug-like molecules and are commonly used to assess and optimise the accuracy of nonbonded parameters in empirical forcefields, and also as a fast-to-compute surrogate of performance for protein-ligand binding free energy estimation. Machine learned potentials (MLPs) show great promise as more accurate alternatives to empirical forcefields, but are not readily decomposed into physically motivated functional forms, which has thus far rendered them incompatible with standard alchemical free energy methods that manipulate individual pairwise interaction terms. However, since the accuracy of free energy calculations is highly sensitive to the forcefield, this is a key area in which MLPs have the potential to address the shortcomings of empirical forcefields. In this work, we introduce an efficient alchemical free energy protocol that enables calculations of rigorous free energy differences in condensed phase systems modelled entirely by MLPs. Using a pretrained, transferrable, alchemically equipped MLP model, we demonstrate sub-chemical accuracy for the solvation free energies of a wide range of organic molecules.

• The paper introduces machine-learned potentials to enable ab initio accuracy in hydration free energy calculations for small molecules.
• It employs a novel soft-core interaction model to stabilize alchemical simulations and reduce energy divergence.
• The scalable framework integrated in OpenMM outperforms traditional force fields by nearly halving mean absolute errors in benchmarks.

Overview of "Computing Hydration Free Energies of Small Molecules with First Principles Accuracy"

The paper presents a methodological advancement in the calculation of hydration free energies for small molecules through the application of machine-learned potentials (MLPs). This work is significant in computational chemistry, particularly within drug discovery, where precise free energy calculations are critical for predicting molecular interactions and properties.

Key Contributions

1. Introduction of MLPs in Free Energy Calculations: The authors have developed an alchemical free energy method compatible with MLPs that enables, for the first time, ab initio accuracy in biomolecular free energy calculations. Traditional force fields, while efficient, lack the precision required for complex biomolecular simulations due to their empirical nature. This research leverages the MACE model, which is pretrained and transferable, to address these limitations.
2. Soft-Core Interaction Model: The paper highlights the use of a novel approach where the training set for MLPs is augmented with dimer configurations that introduce softened two-body interactions. This methodological innovation mitigates issues related to energy divergence encountered when atoms overlap, crucial for stable simulations involving alchemically decoupled systems.
3. Simulation Framework: A scalable and efficient implementation is presented within the OpenMM package, a widely used platform for molecular dynamics simulations. The authors ensure that the approach seamlessly integrates into existing free energy calculation protocols, including pymbar and pmx, broadening its applicability.
4. Benchmarking and Results: Through comprehensive testing, the method demonstrates sub-chemical accuracy in calculating hydration free energies for various organic molecules. The benchmarks reveal significant improvements over classical force fields like GAFF and OpenFF, reducing mean absolute errors by about a factor of two compared to conventional methods.
5. Implications for Drug Discovery: By improving the accuracy of hydration free energy predictions, the method provides a practical tool for drug discovery. Accurate predictions can facilitate the design and optimization of small molecules with better solubility and protein-ligand binding affinities, crucial for therapeutic efficacy.

Implications and Future Directions

The successful application of machine-learned potentials in free energy calculations opens new avenues for computational chemistry, particularly in areas requiring high accuracy and detailed molecular interaction insights, such as drug design and materials science. Future developments could focus on enhancing the scalability of the approach to accommodate larger and more structurally diverse molecules, as well as extending it to more complex chemical environments.

Further exploration and validation of MLP-based free energy calculations in different domains could establish new paradigms for simulations. As the MACE model and its derivatives evolve, their use in combination with traditional force fields may lead to hybrid systems with optimized computational efficiency and accuracy.

Conclusion

This research signifies a step forward in computational chemistry by bridging the gap between empirical force fields and highly accurate quantum mechanical methods. With the capability to calculate hydration free energies with near-first principle precision, this method sets a precedent for future advancements in molecular simulations and their applications in real-world problems, notably in drug discovery and design.