Drug-Protein Interaction Prediction System

About the Project

An AI-powered system for predicting drug-protein interactions using deep learning. It combines molecular analysis, protein feature extraction, and attention-based neural networks to predict binding likelihood between drugs and proteins. The system features a comprehensive web interface for real-time predictions, 3D molecular visualization, and detailed biochemical analysis, designed to assist researchers in drug discovery and development by identifying potential drug targets and assessing binding interactions.

Key Features

• •AttentionDTI model with character-level embeddings and CNN layers
• •Real-time drug-protein binding prediction with confidence scores
• •Interactive 3D visualization of proteins and drug molecules
• •Comprehensive protein analysis with amino acid composition and biochemical properties
• •Molecular descriptor calculation for drugs (LogP, molecular weight, fingerprints)
• •UniProt and AlphaFold database integration for protein structures
• •REST API for programmatic access and batch processing
• •SMILES string processing and validation

Challenges & Solutions

• ⚡Implementing attention mechanisms for cross-modal drug-protein feature fusion
• ⚡Processing and validating complex SMILES strings and protein sequences
• ⚡Integrating multiple databases (UniProt, AlphaFold) for comprehensive analysis
• ⚡Optimizing model performance for real-time inference
• ⚡Balancing model complexity with prediction accuracy and speed

Outcomes & Impact

• ✓Successfully trained AttentionDTI model on Davis, DrugBank, and KIBA datasets
• ✓Deployed dual-interface system (Streamlit UI + FastAPI)
• ✓Achieved real-time predictions with confidence scoring
• ✓Developed at Watson Codefest 2024
• ✓Integrated 25+ protein features and comprehensive molecular descriptors