chaidiscovery/chai-lab; uv pip install chai_lab==0.6.1, current as of 2026-09) under uv run python. Needs Linux, Python >= 3.10 and a CUDA GPU with bfloat16 (A100/H100/L40S recommended; A10/A30/RTX 4090 for smaller complexes). Weights download on first run (relocate with CHAI_DOWNLOADS_DIR). Input: one typed FASTA; MSAs, templates, restraints optional. Apache-2.0 code and weights. Dispatch heavy runs via alterlab-remote-compute.\"\nmetadata:\nskill-author: AlterLab\nversion: \"1.1.0\"\nlast_updated: \"2026-09-23\"Chai-1 (Chai Discovery 2024; chaidiscovery/chai-lab) is an open AlphaFold3-style model\nthat predicts multi-entity biomolecular complexes — proteins, small-molecule ligands, and\nnucleic acids together — from a single typed FASTA. It is particularly used for\nantibody–antigen and protein–ligand complexes, can run with or without MSAs, and accepts\nrestraints to guide the prediction.
Its niche relative to the other folders: one FASTA describing a mixed assembly, and\nantibody–antigen in particular. For a ligand co-fold where you specifically want a binding\naffinity, use alterlab-boltz; for a bare protein, use alterlab-alphafold.
Use this skill when the user wants to:
\n| Scenario | \nUse instead | \n
|---|---|
| Predict a protein–ligand binding affinity | \nalterlab-boltz | \n
| Protein-only or protein–protein folding | \nalterlab-alphafold | \n
| Dock a ligand into a fixed receptor structure | \nalterlab-diffdock | \n
| Look up an experimental complex structure | \nalterlab-pdb | \n
| Design antibody/interface sequences | \nalterlab-proteinmpnn / alterlab-ligandmpnn | \n
Chai-1 reads one FASTA whose records are typed by entity. A protein + ligand example:
\n>protein|name=antibody-Fv\nEVQ...SS\n>protein|name=antigen\nMKT...GG\n>ligand|name=cofactor\nCC(=O)Oc1ccccc1C(=O)O\nchai-lab fold input.fasta out/ # single-sequence, fastest\nchai-lab fold --use-msa-server --use-templates-server input.fasta out/ # recommended\nEach header is <entity_type>|name=<unique label> (the bare type|label form also parses).\nValid entity types: protein, ligand (SMILES), rna, dna, glycan. Names must be\nunique — a repeated name raises. The output directory must be empty.
The common use case: fold an antibody Fv/Fab against its antigen and read the interface\nconfidence (per-model / interface score) to judge whether the predicted epitope/paratope\ncontact is trustworthy. Use restraints when you have partial epitope knowledge.
\n--use-msa-server fetches MSAs\nfrom the shared ColabFold MMseqs2 server (discloseable for sensitive sequences) and\ngenerally improves accuracy at a time cost. Local MSAs are supplied as aligned.pqt\nfiles (convert a3m with chai a3m-to-pqt).--use-templates-server, or your own m8 hit table plus CIFs in\nCHAI_TEMPLATE_CIF_FOLDER.constraint_path, one row per restraint with columns\nrestraint_id, chainA, res_idxA, chainB, res_idxB, connection_type, confidence, min_distance_angstrom, max_distance_angstrom, comment. connection_type is contact\n(residue↔residue) or pocket (chain↔residue, so res_idxA is left blank). Residue\nindices are the residue letter plus its 1-based position (D4), and chains are lettered\nA–Z in input order. Published example: two ground-truth contacts lifted antibody–antigen\ninterface DockQ from ~0.02 to ~0.4 on PDB 7SYZ.A run writes pred.model_idx_{0..4}.cif plus scores.model_idx_N.npz holding\naggregate_score (the ranking number), ptm, iptm, per_chain_ptm,\nper_chain_pair_iptm, has_inter_chain_clashes and chain_chain_clashes. For an\nantibody–antigen job the pair-wise per_chain_pair_iptm entry for the two chains — not the\nglobal score — is what tells you whether the predicted epitope is trustworthy; check the\nclash flags before believing a high score.
The Python entry point is chai_lab.chai1.run_inference(fasta_file=..., output_dir=..., num_diffn_samples=5, num_trunk_recycles=3, seed=..., device=\"cuda:0\"), returning a\nStructureCandidates with cif_paths and ranking_data. Batch an antibody panel against one\nantigen via alterlab-remote-compute (submit → poll → harvest out/).
references/chai_usage.md — install/pinning, FASTA type-tag syntax, MSA/restraint options,\noutputs, and folder-choice guidance. Loaded on demand.Part of the AlterLab Academic Skills suite.
\n","files":[{"path":"evals/evals.json","sizeBytes":3016,"isText":true},{"path":"references/chai_usage.md","sizeBytes":4739,"isText":true},{"path":"SKILL.md","sizeBytes":6046,"isText":true}],"reviewScore":null,"reviewSummary":null,"trust":{"provenance":"trusted-source-unreviewed","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow.","bodySource":null},"bodyLocked":false,"purchaseUrl":null,"sourceUrl":null,"report":{"provenance":"trusted-source-unreviewed","screen":{"ran":true,"outcome":"clean","suspicious":0,"notes":0,"hiddenCharacters":false},"virusScan":{"engine":"clamav","status":"clean","scannedAt":"2026-09-23T18:57:15.874004Z","sha256":"F3C2C809183C7F12E150B3A641475F58CAE5497D7062422501D98A22D499E671","sizeBytes":6560},"review":null,"source":{"repositoryUrl":"https://github.com/AlterLab-IEU/AlterLab-Academic-Skills","path":"skills/bioinformatics/alterlab-chai","license":"MIT","commit":"e4836c08a20da195a11f30f203a8cf23ec30aa95","subtreeSha":"59C48A4432EFAA6233CBB848DD003D5F7A39AF9274DAB505862052F03C55E9B9","lastSyncedAt":"2026-09-23T18:56:52.297238Z"},"reviewedAt":"2026-09-23T18:58:10.423918Z","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow."},"install":[{"target":"skills-cli","command":"npx skills add https://github.com/AlterLab-IEU/AlterLab-Academic-Skills/tree/main/skills/bioinformatics/alterlab-chai"},{"target":"claude-code","command":"claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install alterlab-ieu-alterlab-academic-skills@llmmart"},{"target":"git","command":"git clone https://github.com/AlterLab-IEU/AlterLab-Academic-Skills.git"}]}