kGCN

kGCN applies graph convolutional neural networks to predict molecular properties and compound–protein interactions for cheminformatics and drug discovery.

Key Features:

• Graph convolutional neural networks (GCNs): Represents molecules as graphs and uses GCNs to learn structure–property and structure–interaction relationships.
• Pre-processing: Includes functions to prepare molecular data for model training and evaluation.
• Model tuning (Bayesian optimization): Performs automatic hyperparameter tuning using Bayesian optimization.
• Prediction modes: Supports single-task, multi-task, and multimodal prediction approaches for diverse modeling scenarios.
• Interpretation and explainability: Provides visualization of atomic contributions to predictions to support explainable AI analyses.

Scientific Applications:

• Molecular property and interaction prediction: Predicts molecular properties and compound–protein interactions relevant to cheminformatics studies.
• Compound–protein interaction prediction for matrix metalloproteases: Applied to inhibition assays for MMP-3, MMP-9, MMP-12, and MMP-13 to analyze compound–protein interactions.
• Drug discovery and molecular design: Enables analysis and interpretation of predictions to inform compound optimization and design decisions.

Methodology:

Represents molecular structures as graphs and analyzes them with graph convolutional neural networks, with pre-processing functions, Bayesian optimization for model tuning, and visualization of atomic contributions for interpretation.