Quantification of the Impact of Structure Quality on Predicted Binding Free Energy Accuracy

Relative binding free energy (RBFE) calculations have emerged as a powerful tool in drug discovery, capable of achieving experimental-level accuracy. However, the accuracy is compromised by a multitude of factors, including the initial structure modeling. The current study contributes to the quantification of the impact of initial structure modeling on the accuracy across a diverse set of activity cliff pairs. Along with providing a quantitative relation between the resolution of the crystal structure and free energy accuracy, we also demonstrate the incorporation of a secondary solvation tool (SOLVATE) to increase the free energy accuracy, especially when crystal waters are missing. The study also evaluates the reliability of AI-predicted structures in RBFE calculations, showing their effectiveness in predicting RBFE directionality and assigning nominal resolutions to the predicted structures based on free energy accuracy. These findings provide a set of recommendations for the development of more robust RBFE protocols, informing the use of structural data, solvation techniques, and AI-predicted protein models in drug discovery.