Large Scale Study of Ligand-Protein Relative Binding Free Energy Calculations: Actionable Predictions from...

Area of Science:

• Computational chemistry
• Molecular modeling
• Drug discovery

Background:

• Accurate prediction of protein-ligand binding affinities is vital for drug discovery and personalized medicine.
• Alchemical free energy methods, particularly for relative binding free energy (RBFE) calculations, are important for lead optimization.
• Despite advances, limitations in real-world applicability of these methods persist due to unresolved issues.

Purpose of the Study:

• To provide statistically robust results on the accuracy, precision, and reproducibility of RBFE calculations.
• To evaluate ensemble-based methods, specifically Thermodynamic Integration with Enhanced Sampling (TIES), for reliable uncertainty quantification.
• To investigate the impact of various alchemical factors on RBFE prediction accuracy.

Main Methods:

• Utilized a large dataset of over 500 ligand transformations across 14 protein targets.
• Employed ensemble-based methods, including TIES (Thermodynamic Integration with Enhanced Sampling), for chaotic molecular dynamics.
• Analyzed statistical distributions of free energy calculations and assessed factors like ligand charge method, flexibility, and alchemical region size.

Main Results:

• RBFE calculations achieved chemical accuracy in all tested cases.
• Ensemble simulations revealed non-normal statistical distributions of free energy calculations.
• The replica exchange with solute tempering method was found to degrade RBFE predictions.

Conclusions:

• TIES provides statistically robust and chemically accurate RBFE calculations with reliable uncertainty quantification.
• Recommendations are provided for the reliable application of RBFE methods, addressing factors influencing prediction accuracy.
• Findings contribute to enhancing the applicability of alchemical free energy methods in drug discovery and personalized medicine.