Extended connectivity interaction features: improving binding affinity prediction through chemical description.

Area of Science:

• Computational chemistry
• Structural biology
• Drug discovery

Background:

• Machine-learning scoring functions (SFs) show promise for predicting protein-ligand binding affinity.
• Current methods often focus on complex algorithms rather than fully exploiting chemical descriptions.

Purpose of the Study:

• To introduce Extended Connectivity Interaction Features (ECIF) for describing protein-ligand complexes.
• To develop and evaluate machine-learning SFs using ECIF for improved binding affinity prediction.

Main Methods:

• ECIF were developed as atom-type pair counts considering atomic connectivity.
• Machine-learning models were built using ECIF to predict protein-ligand affinities (pKd/pKi).
• Model performance was assessed using the Comparative Assessment of Scoring Functions 2016 benchmark.

Main Results:

• Models utilizing ECIF achieved high predictive power.
• A Pearson correlation coefficient of 0.857 was obtained using ECIF alone.
• Combining ECIF with ligand descriptors yielded a correlation coefficient of 0.866.

Conclusions:

• ECIF effectively describe protein-ligand complexes for enhanced binding affinity prediction.
• The developed machine-learning SFs demonstrate significant improvements over existing methods.
• The ECIF approach offers a valuable tool for drug discovery and computational chemistry research.