Integrating Machine Learning Interatomic Potentials with MMPBSA for Accurate Protein-Ligand Binding Free Energy...

Area of Science:

• Computational chemistry
• Molecular modeling
• Drug discovery

Background:

• End-point binding free energy (BFE) methods like MMPBSA are crucial for estimating protein-ligand binding affinity.
• Their accuracy is limited by statistical approximations and classical potential energy surfaces (PES).

Purpose of the Study:

• To develop a novel hybrid framework, AIQM-PBSA, overcoming the PES bottleneck in BFE calculations.
• To enhance the accuracy and reliability of protein-ligand binding affinity predictions.

Main Methods:

• Developed AIQM-PBSA, integrating the ONIOM scheme with the PBSA model.
• Employed the AIQM3 machine learning interatomic potential (MLIP) to refine molecular mechanics (MM) energy terms.
• Evaluated solvation contributions using the PBSA formalism.

Main Results:

• AIQM-PBSA demonstrated substantial improvements in predictive accuracy across diverse protein-ligand systems.
• Achieved high Pearson R values (0.84, 0.82) on benchmark datasets and significant correlations on the Schrödinger JACS set.
• Outperformed traditional MMPBSA and ANI-2x by utilizing MLIPs for gas-phase interaction energies.

Conclusions:

• AIQM-PBSA offers a robust and generalizable framework for biomolecular recognition.
• Leverages advanced MLIPs to achieve quantum mechanical (QM) accuracy with high computational efficiency.
• Significantly improves the reliability of end-point free energy calculations.