Boltz

Boltz Structure Prediction

Predict biomolecular structures using Boltz, an open-source diffusion model. Boltz handles proteins, RNA, DNA, small molecules, ions, and covalent modifications in a single model without requiring multiple sequence alignments (MSA-optional). It serves as a strong open-source alternative to AlphaFold3.

Requirements

• Python 3.10+
• 24 GB GPU VRAM minimum (A10G/A100 recommended)
• ~10 GB disk for model weights

Installation

pip install boltz

Input Format (YAML)

Boltz uses YAML for flexible entity specification:

# complex.yaml — protein + ligand
version: 1
sequences:
  - protein:
      id: A
      sequence: MTEYKLVVVGAGGVGKSALTIQLIQNHFVDEYDPTIEDSY...
  - ligand:
      id: B
      smiles: "CC1=CC=C(C=C1)S(=O)(=O)N"  # or CCD code
      ccd: ATP  # alternative: use CCD code

# binder-target complex
version: 1
sequences:
  - protein:
      id: [A, B]  # homodimer
      sequence: MTEYKLVVVGAGGVGKS...
      count: 2
  - protein:
      id: C
      sequence: EVQLVESGGGLVQPGG...  # binder

Running Predictions

# Single prediction
boltz predict complex.yaml \
    --out_dir results/ \
    --accelerator gpu \
    --devices 1 \
    --num_workers 4

# Batch prediction (multiple YAML files)
boltz predict inputs/ \
    --out_dir results/ \
    --accelerator gpu

# Without MSA (faster, slightly lower accuracy for monomers)
boltz predict complex.yaml \
    --out_dir results/ \
    --use_msa_server false

Python API

from boltz.main import predict

predict(
    data="complex.yaml",
    out_dir="results/",
    accelerator="gpu",
    devices=1,
    num_predictions=1,  # ensemble size
    recycling_steps=3,
    diffusion_samples=1
)

Output Files

results/
  boltz_results_complex/
    predictions/
      complex/
        complex_model_0.cif          # Predicted structure (CIF format)
        complex_confidence_model_0.json  # Confidence scores
    lightning_logs/                  # Training logs (ignore)

Confidence Metrics

import json

with open("complex_confidence_model_0.json") as f:
    conf = json.load(f)

# Key metrics
plddt = conf["plddt"]                    # Per-residue confidence (0-100)
ptm = conf["ptm"]                        # Global fold confidence (0-1)
iptm = conf["iptm"]                      # Interface confidence (0-1)
ligand_iptm = conf.get("ligand_iptm")    # Ligand interface confidence
pde = conf.get("pde")                    # Predicted Distance Error

print(f"pTM={ptm:.3f}, ipTM={iptm:.3f}")

Quality Thresholds

vs. AlphaFold2/3

Convert CIF to PDB

# Using BioPython
python3 -c "
from Bio.PDB import MMCIFParser, PDBIO
parser = MMCIFParser()
structure = parser.get_structure('pred', 'complex_model_0.cif')
io = PDBIO()
io.set_structure(structure)
io.save('complex_model_0.pdb')
"