A knowledge-guided strategy for improving the accuracy of scoring functions in binding affinity prediction

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Current scoring functions struggle with accurate protein-ligand binding affinity prediction.
• Structure-based drug design relies heavily on these scoring functions.
• Existing methods often lack the precision needed for reliable predictions.

Purpose of the Study:

• To develop a novel knowledge-guided scoring (KGS) strategy for enhanced binding affinity prediction.
• To improve the accuracy of absolute binding constant calculations.
• To provide a practical solution for the limitations of current scoring functions.

Main Methods:

• Developed an automatic algorithm to summarize key protein-ligand interactions as pharmacophore models.
• Implemented a strategy to identify reference complexes with maximal similarity to query complexes.
• Evaluated the KGS strategy with X-Score and PLP on HIV protease, carbonic anhydrase, and trypsin datasets.

Main Results:

• The KGS strategy demonstrated more accurate predictions compared to standalone scoring functions.
• Improved performance was observed on both crystal structures and docking poses.
• Effectiveness was particularly notable when X-Score or PLP alone showed poor performance.

Conclusions:

• The KGS strategy offers a general approach without requiring re-parameterization of existing scoring methods.
• Its accuracy is theoretically proportional to the growing body of experimental binding data.
• KGS presents a practical advancement for improving the accuracy of protein-ligand binding affinity predictions.