Global free energy scoring functions based on distance-dependent atom-type pair descriptors

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Protein-ligand docking scoring functions aim to predict binding affinity.
• Current scoring functions show limited correlation between predicted and observed binding energies.
• Predicting binding free energies from crystal structures remains a challenge.

Purpose of the Study:

• To evaluate the prediction of binding free energies using quantitative structure-activity relationship (QSAR) techniques in a proteochemometric approach.
• To develop novel protein-ligand interaction descriptors for improved scoring functions.
• To compare the performance of new scoring functions against existing methods.

Main Methods:

• Utilized a subset of the PDBbind09-CN refined dataset (N=1387).
• Developed new distance-binned Crippen-like atom pair descriptors for protein-ligand interactions.
• Applied QSAR and proteochemometric modeling to generate scoring functions.
• Validated scoring functions using cross-validation and external test sets.

Main Results:

• Developed scoring functions with good predictive performance on a diverse dataset (R(2)(out-of-bag) = 0.48, R(p) = 0.69).
• Achieved significantly better performance compared to existing scoring functions on large, diverse test sets.
• Demonstrated that ligand-only descriptors can yield competitive or superior scoring functions for certain protein families.

Conclusions:

• Novel interaction descriptors and proteochemometric approaches can significantly improve binding free energy prediction.
• Ligand-based descriptors offer a viable alternative for scoring function development in specific cases.
• A standardized validation protocol is crucial for future scoring function development and comparison.