QuantumBind-RBFE: Accurate Relative Binding Free Energy Calculations Using Neural Network Potentials.

Area of Science:

• Computational chemistry
• Drug discovery and development

Background:

• Accurate prediction of protein-ligand binding affinities is critical for efficient drug discovery.
• Existing ligand force fields present limitations impacting prediction accuracy.
• Neural network potentials (NNPs) offer a promising alternative for enhanced accuracy.

Purpose of the Study:

• To validate the accuracy of relative binding free energy (RBFE) predictions using a novel NNP model, AceFF 1.0.
• To assess the performance of AceFF 1.0 against established methods like GAFF2 and ANI2-x.
• To evaluate the computational efficiency and applicability of AceFF 1.0 for diverse drug-like molecules.

Main Methods:

• Development and utilization of AceFF 1.0, a TensorNet-based NNP model for small molecules.
• Validation using established benchmarks for binding affinity prediction.
• Comparative analysis against GAFF2 (molecular mechanics) and ANI2-x (NNPs).
• Assessment of simulation speed using a 2 fs time step.

Main Results:

• AceFF 1.0 demonstrates improved accuracy and correlation in binding affinity predictions compared to GAFF2 and ANI2-x.
• The model shows comparable correlations to OPLS4, with slightly lower accuracy.
• NNP simulations with AceFF 1.0 can be run at a 2 fs time step, offering significant speed improvements.
• The model supports diverse drug-like compounds, including charged molecules.

Conclusions:

• AceFF 1.0 shows significant promise for advancing free energy calculations in drug discovery.
• The current generation of AceFF 1.0 is already practical for use in research.
• The code and NNP model are publicly available, facilitating further research and development.