ECIF

ECIF encodes protein–ligand interactions as connectivity-based atom-type pair counts used as descriptors for machine-learning prediction of binding affinity (pKd/pKi).

Key Features:

• Atom-Type Pair Counts: Connectivity-based counts of atom-type pairs between protein and ligand atoms that encode each atom's connectivity within the complex.
• Connectivity Consideration: Explicit incorporation of atom connectivity to define specific atom-type pair types for more granular interaction description.
• Chemical Description: A novel connectivity-based chemical description of protein–ligand interactions as feature vectors.
• Machine-Learning Integration: Designed to serve as input descriptors for machine-learning scoring functions predicting binding affinity (pKd/pKi).
• Compatibility with Ligand Descriptors: Can be combined with ligand descriptors to enhance predictive performance.

Scientific Applications:

• Machine-Learning Scoring Functions: Construction of ML models for predicting binding affinities (pKd/pKi) using ECIF descriptors.
• Comparative Assessment (CASF-2016): Evaluation in CASF-2016 yielded Pearson correlations of 0.857 when used alone and 0.866 when combined with ligand descriptors, demonstrating improved performance relative to conventional scoring functions.
• Drug Discovery: Improved prediction of binding affinities relevant to ligand optimization and drug development projects.

Methodology:

Compute connectivity-based atom-type pair counts between protein and ligand atoms and use these counts as feature vectors in machine-learning models to predict binding affinity (pKd/pKi); performance reported via Pearson correlation in CASF-2016.