Simple Workflow for Ligand Binder Design

Developed at Artificial Intelligence Protein Design Lab

This repository provides a reproducible workflow for ligand binder design using diffusion models. The workflow integrates multiple state-of-the-art tools to generate and validate protein binders for small molecule ligands.

Workflow Steps

The workflow consists of 9 sequential steps:

1. Parameters Setup (0_params/) - Ligand preparation and parameter files
2. Backbone Generation (1_diffusion/) - Structure generation using RFdiffusion
3. Backbone Filtering (2_backbone_filter/) - DSSP and SASA-based filtering
4. Sequence Design (3_lmpnn/) - Sequence generation using LigandMPNN
5. Rosetta Scoring (4_rscore_filter/) - Energy-based filtering
6. AlphaFold3 Prediction (5_af3/) - Structure prediction and validation
7. Boltz Prediction (6_boltz/) - Alternative structure prediction
8. PLACER Analysis (7_placer/) - Binding site analysis
9. RMSD Filtering (8_rmsd_filter/) - Final structure validation

Features

• Reproducible Workflow: Complete protocol from ligand input to validated binders
• Multiple Validation Steps: Combines geometric, energetic, and structural filters
• Modern AI Tools: Utilizes RFdiffusion, LigandMPNN, AlphaFold3, and Boltz
• Scalable: Designed for SLURM-based cluster environments
• Performance Optimized: Multiprocessing enabled for DSSP/SASA/Rosetta scoring and RMSD calculations
• Enhanced RMSD Analysis: Biopython-based structure handling with RDKit for ligand symmetry-aware RMSD calculations

Requirements

• RFdiffusion All-Atom
• LigandMPNN
• Rosetta
• AlphaFold3
• Boltz
• PLACER
• TMalign
• PyMOL
• Python 3.8+
• PyArrow (for parquet file handling)

Note: This workflow uses parquet file format to handle double headers efficiently. Parquet files can be viewed and analyzed using VSCode's Data Wrangler extension.

Usage

1. Place your ligand files in the 0_params/ directory
2. Run each workflow step sequentially using the provided run.sh scripts
3. Review and filter results at each step as needed
4. Final candidates will be available in 8_rmsd_filter/