---

name: restraints-glycans description: "Author, validate, and troubleshoot Chai-1 restraint CSVs, covalent bonds, and glycan inputs." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Chai-1 Restraints and Glycans

Use this sub-skill when a task involves Chai-1 contact restraints, pocket restraints, covalent bond restraints, glycan FASTA records, or restraint CSV validation.

Start Here

1. Read references/restraints-and-glycans.md for the CSV schema, contact/pocket/covalent semantics, residue/atom notation, chain naming, and glycan syntax.
2. Read references/api-reference.md when using chai_lab.chai1.run_inference, parse_pairwise_table, write_pairwise_table, or PairwiseInteraction directly.
3. Run scripts/validate_restraints.py --help and use the helper before launching inference.
4. Read references/troubleshooting.md when parsing succeeds but restraints do not affect the result, covalent atom lookup fails, or glycan syntax is rejected.

Typical Workflow

• Confirm the FASTA chain naming mode first: automatic Chai chains are A, B, C, ... by FASTA order unless inference uses fasta_names_as_cif_chains=True.
• Create a restraint CSV with unique restraint_id values and connection_type set to contact, pocket, or covalent.
• Use residue notation such as A219, residue-atom notation such as A219@CA, and atom-only notation such as @C1 only for covalent ligand/glycan sides where Chai should target the first token/residue.
• Validate cheap schema, notation, chain-name, glycan-string, and simple residue-position errors with python scripts/validate_restraints.py restraints.csv --fasta input.fasta.
• Pass the CSV through constraint_path in Python or the corresponding chai-lab fold --constraint-path CLI option.

Boundaries

• For general FASTA entity syntax, modified residues, ligands, DNA/RNA, and entity-name uniqueness, use ../input-data-formats/SKILL.md.
• For MSA directories, .aligned.pqt, ColabFold server use, and template hit files, use ../msa-templates/SKILL.md.
• For full folding commands, output directories, devices, seeds, samples, and StructureCandidates, use ../cli-inference/SKILL.md.