Drug Repurposing Research

AI-driven computational drug repurposing using multi-modal frameworks

Project Overview

Led the development of three complementary AI frameworks for computational drug repurposing at HUST VLR Lab, focusing on identifying novel therapeutic applications for existing pharmaceuticals while reducing traditional development timelines.

Research Framework

KIDE (Knowledge-enhanced Intelligent Drug Evaluation)

  • Multi-agent System: Sophisticated framework integrating biomedical knowledge graphs with LLM-based analysis
  • Knowledge Integration: Combined data from 17 biomedical repositories with 37,000+ drug entities
  • Intelligent Analysis: Advanced prompt engineering for mechanism analysis and therapeutic potential assessment

HyResBio (Hybrid Residual Biological Networks)

  • Dual-network Approach: Combined Hybrid Graph Neural Networks (HyGNN) and Residual Biological Networks (ResBioNet)
  • Enhanced Prediction: Improved accuracy and biological interpretability for drug-target interactions
  • Data Augmentation: Implemented SMOTE and GAN-based techniques for handling imbalanced datasets

MSDR (Multi-Expert System for Drug Repurposing)

  • Deep Learning Integration: Leveraged deep neural networks and Langchain architecture
  • Molecular Analysis: Advanced methods for analyzing molecular structures and predicting therapeutic potential
  • Multi-dimensional Scoring: Combined knowledge graph embedding, mechanism analysis, and safety evaluation

Technical Achievements

Database Construction

  • Comprehensive Integration: First OA-specific fusion database with 37,000+ drug entities
  • Specialized Modules: Disease-specific knowledge modules for targeted analysis
  • Quality Control: Advanced data validation and quality monitoring systems

Algorithm Innovation

  • Multi-modal Fusion: Successfully integrated knowledge graphs, LLMs, and GNNs
  • Data Handling: Advanced methods for imbalanced dataset management
  • Interpretability: Designed explainable AI frameworks for drug mechanism analysis

Research Impact

  • Manuscripts: Preparing three research manuscripts for high-impact journals
  • Collaboration: Working with medical collaborators for pharmacological validation
  • Methodology: Established “data integration - intelligent modeling - mechanism analysis” framework

References