3D-RISM-AI: A Machine Learning Approach to Predict Protein-Ligand Binding Affinity Using 3D-RISM

Area of Science:

• Computational chemistry
• Biophysics
• Machine learning

Background:

• Hydration free energy (HFE) is crucial for accurate protein-ligand binding free energy (BFE) prediction.
• Existing methods using three-dimensional reference interaction model (3D-RISM) for HFE alone do not correlate well with experimental BFE.
• Improved computational methods are needed for reliable BFE prediction in drug discovery.

Purpose of the Study:

• To develop a novel machine learning approach, 3D-RISM-AI, for predicting protein-ligand binding free energy (BFE).
• To integrate hydration free energy (HFE) calculated via 3D-RISM into a machine learning model.
• To enhance the accuracy of BFE predictions for diverse protein-ligand complexes.

Main Methods:

• Utilized the PDBbind database (ver. 2018) comprising approximately 4000 protein-ligand complexes.
• Incorporated structural metrics, intra-/intermolecular energies, and 3D-RISM-derived HFEs into the machine learning model.
• Trained the 3D-RISM-AI model to predict experimental BFE values.

Main Results:

• The 3D-RISM-AI model achieved a strong correlation with experimental BFE data (Pearson's correlation coefficient = 0.80).
• The model demonstrated a low root-mean-square error of 1.91 kcal/mol, indicating high prediction accuracy.
• Key predictive factors identified include changes in solvent accessible surface area and ligand hydration properties.

Conclusions:

• The 3D-RISM-AI approach effectively predicts protein-ligand binding free energies.
• Integrating 3D-RISM-calculated hydration free energy with machine learning significantly enhances prediction accuracy.
• This method offers a promising tool for computational drug discovery and molecular modeling.