Overview

This project focuses on developing an AI-powered pipeline for drug discovery by integrating molecular docking simulations, knowledge graph reasoning, and explainable AI (XAI) techniques. Our goal is to accelerate lead compound identification and provide interpretable insights into protein-ligand interactions, enabling researchers to understand why certain drug candidates are predicted to be effective.

Key Objectives

• Build an automated molecular docking framework powered by deep learning models.
• Construct a domain-specific biomedical knowledge graph for drug-target relationships.
• Integrate graph reasoning and link prediction for novel drug-target interaction discovery.
• Develop XAI methods to provide human-understandable explanations for docking results and predictions.

Methods & Approach

• Molecular Docking: Use classical docking tools (AutoDock Vina, RDKit) enhanced by graph neural networks (GNNs) to predict binding affinities and pose ranking.
• Knowledge Graph: Construct a large-scale drug-target-pathway knowledge graph using biomedical databases (DrugBank, ChEMBL).
• Reasoning: Employ graph-based algorithms (e.g., R-GCN, TransE) for link prediction and candidate discovery.
• XAI: Apply feature attribution (SHAP, Integrated Gradients) and attention-based visualization to interpret molecular features driving predictions.
• Evaluation: Benchmark against known datasets (PDBBind, BindingDB) and compare with traditional docking approaches.

Current Progress

Applications

• Accelerated lead compound screening for drug discovery.
• Discovery of novel drug-target interactions using graph reasoning.
• Interpretable drug design to support medicinal chemistry and pharmacology research.

Team

• Principal Investigator: Chulwoo Pack
• Graduate Students:
• Collaborators:

Related Publications

• See related works:

Resources

• GitHub Repository
• Knowledge Graph Visualization
• Docking Pipeline Documentation