| name | chai1-structure-prediction |
| description | Run pinned Chai-1 inference for proteins, complexes, nucleic acids, ligands, covalent bonds, templates, or restraints. Use when multimolecular structure prediction needs Chai-1-specific inputs and confidence outputs. |
| license | MIT |
Chai-1 Structure Prediction
Gate
Ask once before installing a pinned chai_lab release, downloading weights,
using GPU compute, or contacting public MSA/template servers. Name each remote
host and never upload private sequences or structures without approval. Follow
$cx-compute-environment.
Workflow
- Define every entity and state: protein/RNA/DNA sequence, ligand SMILES/CCD,
modifications, stoichiometry, covalent bonds, restraints, and templates.
- Pin Chai-1 version/commit, weight checksum, input schema, MSA/template mode,
number of samples, seeds, hardware, precision, and all restraints.
- Smoke-test
chai-lab fold --help and an upstream example. Run the validated
FASTA/context into artifacts/<run-id>/chai1/; preserve inputs, MSAs,
templates, restraint files, structures, scores, logs, and failures.
- Separate model confidence, interface confidence, geometry checks, and any
downstream affinity estimate. Compare samples and inspect clashes, ligand
state, covalent geometry, low-confidence regions, and restraint satisfaction.
- Record with
$science-provenance; use $science-review before conclusions.
Boundaries
- Restraint satisfaction is not independent evidence when restraints were input.
- A plausible complex or ligand pose is not proof of binding or affinity.
- Report results from public ColabFold-style MSAs separately from local MSAs.