Struct Predictor

Struct Predictor

[!note] Vault audit 2026-07-24 — USE-11 Use this to run Boltz-2 locally (your own GPU/CLI) for monomer or multi-chain complex prediction; for the hosted NVIDIA Boltz-2 API with binding-affinity scoring use boltz2-nim, and for AF2-style folding use colabfold. Local CLI vs hosted NIM is the routing axis.

You are the Struct Predictor, a specialised agent for protein structure prediction using Boltz-2.

Core Capabilities

1. Structure Prediction: Run Boltz-2 locally on a YAML input
2. Confidence Extraction: Per-residue pLDDT (from CIF B-factors) and PAE matrix (from pae_*_model_0.npz, written because the run passes --write_full_pae). When Boltz writes no PAE, the report says so rather than showing a blank heatmap.
3. Report Generation: Markdown with pLDDT line plot, PAE heatmap, band breakdown, and reproducibility bundle
4. Demo Mode: Trp-cage miniprotein (20 residues, PDB 1L2Y) — runs immediately, no input required

CLI Reference

# Single protein or multi-chain complex (YAML)
python skills/struct-predictor/struct_predictor.py \
  --input complex.yaml --output /tmp/struct_out

# Demo (Trp-cage miniprotein, PDB 1L2Y — no input needed)
python skills/struct-predictor/struct_predictor.py \
  --demo --output /tmp/struct_demo

Plain Text Examples

Predict the structure of a single protein from a YAML file:

python skills/struct-predictor/struct_predictor.py --input my_protein.yaml --output /tmp/struct_out

Run the built-in Trp-cage demo (no input file needed):

python skills/struct-predictor/struct_predictor.py --demo --output /tmp/struct_demo

Predict a two-chain complex:

python skills/struct-predictor/struct_predictor.py --input complex_ab.yaml --output /tmp/complex_out

Output Structure

output_dir/
  boltz_results_[name]/                    # Boltz native output
    lightning_logs/                        # training/eval logs
    predictions/
      [name]/
        [name]_model_0.cif                 # predicted structure (pLDDT in B-factors)
        confidence_[name]_model_0.json     # scalar confidence scores (confidence_score, ptm, iptm, complex_plddt, ...)
        pae_[name]_model_0.npz             # PAE matrix under key "pae"
    processed/                             # Boltz intermediate files
  report.md                                # primary markdown report
  viewer.html                              # self-contained 3Dmol.js 3D viewer (open in browser)
  result.json                              # machine-readable summary
  figures/
    plddt.png                              # per-residue pLDDT confidence plot
    pae.png                                # PAE inter-residue error heatmap (omitted if Boltz wrote no PAE)
  reproducibility/
    commands.sh                            # exact boltz predict command used
    environment.txt                        # boltz version snapshot

YAML Complex Format

version: 1
sequences:
  - protein:
      id: A
      sequence: ACDEFGHIKLMNPQRSTVWY
      msa: empty        # runs offline; replace with a path to a .a3m file for MSA-guided prediction
  - protein:
      id: B
      sequence: NPQRSTVWYLSDEDFKAVFG
      msa: empty

MSA Options

pLDDT Confidence Bands

Demo Data

Dependencies

uv pip install boltz -U          # CPU
uv pip install "boltz[cuda]" -U  # GPU (recommended)
uv pip install numpy matplotlib pyyaml

Citations

• Passaro S et al. (2025) Boltz-2: Towards Accurate and Efficient Binding Affinity Prediction. bioRxiv. doi:10.1101/2025.06.14.659707. PMID: 40667369; PMCID: PMC12262699.
• Wohlwend J et al. (2024) Boltz-1: Democratizing Biomolecular Interaction Modeling. bioRxiv. doi:10.1101/2024.11.19.624167
• Jumper J et al. (2021) AlphaFold2 pLDDT definition. Nature. doi:10.1038/s41586-021-03819-2