Boltz-2 Protein-Ligand Complex Prediction

Predicts the structure and binding affinity of protein-ligand complexes using the Boltz-2 model. Supports both separate protein/ligand inputs and pre-formed complex PDBs.

Quick Start

1. Select input mode: separate components or complex PDB.
2. Provide protein (A3M or FASTA) and ligand (SMILES or CCD) or upload a complex PDB.
3. Set prediction parameters as needed.
4. Run the node to obtain predicted structures, confidence, and affinity scores.

Setup Guide

1. Prepare Inputs

1. Choose input mode: separate_components (default) or complex_pdb.
2. For separate_components, provide:
3. Protein sequence (A3M or FASTA)
4. Ligand (SMILES string or CCD code)
5. For complex_pdb, upload a pre-formed protein-ligand complex PDB.

2. Configure Prediction

1. Adjust optional parameters (recycling steps, diffusion samples, output format, etc.) as needed.
2. Enable or disable binding affinity prediction.

Basic Usage

Predicting Protein-Ligand Complex Structure

• Use separate_components mode for de novo prediction from sequence/MSA and ligand.
• Use complex_pdb mode to refine or score an existing complex.
• Adjust recycling_steps for quality vs. speed.
• Set diffusion_samples >1 for diverse predictions.

Configuration

Required Inputs

Field	Description	Type	Example
input_mode	Input mode: separate components or PDB	STRING	separate_components
seed	Random seed for reproducibility	INT	42

Optional Inputs

Field	Description	Type	Example
protein_input_type	Protein input type: a3m or fasta	STRING	a3m
protein_a3m	Protein MSA (single-entry dict)	A3M	{"A": "..."}
protein_fasta	Protein FASTA sequence	FASTA	">A\nMKT..."
ligand_type	Ligand input type: smiles or ccd	STRING	smiles
ligand_smiles	Ligand SMILES string	STRING	CCO
ligand_ccd	Ligand CCD code	STRING	ATP
complex_pdb	Pre-formed protein-ligand complex	PDB	{"complex": "..."}
recycling_steps	Number of recycling iterations	INT	3
diffusion_samples	Number of structure samples to generate	INT	1
predict_affinity	Predict binding affinity	BOOLEAN	True
use_potentials	Use potentials for improved quality	BOOLEAN	False
output_format	Output format: pdb or mmcif	STRING	pdb

Outputs

Field	Description	Example
structures.pdb	Predicted protein-ligand complex structures	{"A": "..."}
confidence.json	Confidence scores for predictions	{"A": 0.92}
affinity.json	Predicted binding affinity metrics	{"A": -7.1}

Best Practices

Input Preparation

• Use high-quality MSA (A3M) for best structure prediction results.
• Provide chemically valid SMILES or correct CCD codes for ligands.

Parameter Tuning

• Increase recycling_steps for higher accuracy (at the cost of speed).
• Use multiple diffusion_samples to explore structural diversity.

Troubleshooting

Common Issues

• Missing or invalid input: Ensure all required fields are provided and correctly formatted.
• Low confidence or affinity scores: Try increasing recycling steps or providing better MSA/ligand data.
• Affinity prediction fails: Confirm ligand is present and correctly specified.

Need Help?

• Contact support for further assistance.