AI-Enhanced Virtual Screening Pipeline

Target Identification

This program develops a generalizable computational pipeline for rapid virtual screening against emerging drug targets. The pipeline leverages deep learning models trained on protein-ligand interaction data to predict binding affinities with high accuracy.

Approach

Our pipeline integrates multiple AI/ML approaches:

• Protein structure prediction: AlphaFold2-based target structure modeling
• Pocket detection: Deep learning-based binding site identification
• Molecular generation: Generative models for novel compound design
• Affinity prediction: Graph neural networks for binding affinity estimation
• ADMET modeling: Multi-task learning for ADMET property prediction

Current Status

The platform is in the target stage, with initial validation against known drug-target pairs showing promising enrichment factors.

Key Milestones

• Framework design: Modular pipeline architecture defined
• Model training: Initial models trained on PDBbind dataset
• Benchmark validation: Retrospective validation on known targets underway