protenix

name: protenix description: > Structure prediction using Protenix, ByteDance's open-source PyTorch reproduction of AlphaFold 3. Use this skill when: (1) Predicting protein/DNA/RNA/ligand/ion complex structures, (2) Need AF3-level accuracy with open-source code, (3) MSA-free fast prediction (--no-use-msa), (4) Multi-seed ensemble predictions, (5) Alternative to Chai or Boltz for validation.

For QC thresholds, use protein-qc. For Chai prediction, use chai. For Boltz prediction, use boltz. license: MIT category: design-tools tags: [structure-prediction, validation, af3, open-source] source: https://github.com/hgbrian/biomodals

Protenix Structure Prediction

Protenix is an open-source PyTorch reproduction of AlphaFold 3 by ByteDance. Supports: protein, DNA, RNA, ligand (SMILES), ion.

Prerequisites

How to run

FASTA input (simplest)

# Single protein
echo ">protein|A
MAWTPLLLLLLSHCTGSLSQPVLTQPTSLSASPGASARFTCTLRSGINVGTYRIYWYYQQKPGSLP" > test.faa

modal run modal_protenix.py --input-faa test.faa

Protein complex

cat > complex.faa << 'EOF'
>protein|A
MKTAYIAKQRQISFVKSHFSRQLERRLEQLKQLEQQ...
>protein|B
MVLSPADKTNVKAAWGKVGAHAGEYGAEALERMFL...
EOF

modal run modal_protenix.py --input-faa complex.faa --seeds "42,43,44"

With ligand (SMILES)

cat > protein_ligand.faa << 'EOF'
>protein|A
MKTAYIAKQRQISFVKSHFSRQLE...
>ligand|caffeine
CN1C=NC2=C1C(=O)N(C(=O)N2C)C
EOF

modal run modal_protenix.py --input-faa protein_ligand.faa

With MSA (higher accuracy, slower)

modal run modal_protenix.py --input-faa complex.faa --use-msa

JSON input (native Protenix format)

[
    {
        "name": "my_complex",
        "sequences": [
            {"proteinChain": {"sequence": "MKTAYIAKQRQISFVK...", "count": 1}},
            {"proteinChain": {"sequence": "MVLSPADKTNVKAA...", "count": 1}}
        ]
    }
]

modal run modal_protenix.py --input-json complex.json

Key parameters

Supported entity types

Output format

output/
├── predictions/
│   ├── my_complex_seed-42_sample-0.cif    # Best model
│   ├── my_complex_seed-42_sample-1.cif
│   └── confidence_my_complex_seed-42.json # pLDDT, pTM, ipTM

Protenix vs Chai vs Boltz

Typical performance

Troubleshooting

Next: protein-qc for filtering and ranking.