Geometric graph learning with extended atom-types features for protein-ligand binding affinity prediction.

Area of Science:

• Computational chemistry and cheminformatics
• Machine learning in drug discovery
• Molecular modeling and simulation

Background:

• Accurate prediction of protein-ligand binding affinity is crucial for efficient drug design.
• Machine learning (ML) methods are increasingly utilized due to their accuracy and the growing availability of data.
• Graph theory provides a natural framework for modeling molecular interactions.

Purpose of the Study:

• To enhance graph-based machine learning models for protein-ligand interaction studies.
• To integrate extensive atom types, specifically SYBYL and Extended Connectivity Interactive Features (ECIF), into multiscale weighted colored graphs (MWCG).
• To develop and validate new predictive models for binding affinity.

Main Methods:

• Integration of SYBYL and ECIF atom types into multiscale weighted colored graphs (MWCG).
• Application of the gradient boosting decision tree (GBDT) machine learning algorithm.
• Development of two models: ^sybylGGL-Score and ^ecifGGL-Score.
• Validation using CASF-2007, CASF-2013, and CASF-2016 benchmark datasets.

Main Results:

• Both ^sybylGGL-Score and ^ecifGGL-Score achieved state-of-the-art results on benchmark datasets.
• The ^sybylGGL-Score model demonstrated superior performance compared to other state-of-the-art methods across all benchmarks.
• The best-performing SYBYL atom-type model was further validated on independent test sets.

Conclusions:

• The proposed graph-based machine learning approach significantly improves the prediction of protein-ligand binding affinity.
• The SYBYL atom-type model represents a substantial advancement in binding affinity prediction accuracy for drug design.
• These models offer a powerful tool for accelerating the drug discovery process.