Accurate Hydration Free Energy Calculations for Diverse Organic Molecules With a Machine Learning Force Field.

Area of Science:

• Computational Chemistry
• Drug Discovery
• Machine Learning

Background:

• Classical force fields limit accuracy in free energy perturbation (FEP) calculations for drug discovery.
• Machine learning force fields (MLFFs) offer a way to achieve quantum mechanical accuracy at lower computational cost than ab initio molecular dynamics (AIMD).
• Systematic protocols and benchmarking are lacking for MLFFs in FEP calculations.

Purpose of the Study:

• To develop a general and robust workflow for hydration free energy (HFE) calculations using MLFFs.
• To benchmark the performance of MLFFs against classical force fields and DFT-based models for HFE prediction.
• To advance the application of MLFFs in predicting thermodynamic properties for drug discovery.

Main Methods:

• Developed a workflow for HFE calculations independent of MLFF architecture.
• Utilized a broadly trained MLFF, Organic_MPNICE.
• Employed the solute-tempering technique for enhanced statistical and conformational sampling.

Main Results:

• Achieved sub-kcal/mol average errors in HFE predictions compared to experimental data.
• Demonstrated superior performance over state-of-the-art classical force fields and DFT-based implicit solvation models.
• Validated the workflow on a diverse set of 59 organic molecules.

Conclusions:

• The presented workflow enables accurate, ab initio-quality HFE predictions using MLFFs.
• This approach significantly advances the use of MLFFs in computational drug discovery and thermodynamic property prediction.