Chai-1

Predict the structure of complexes containing proteins, nucleic acids, and small molecules.

API/CLI tool ID: chai

The input fields depend on the value of inputFormat.

Understanding inputs and outputs

Complex prediction calculates how multiple molecules are arranged together. In molecules input, specify each molecule’s type and chain separately. Proteins and nucleic acids use sequences; small molecules use SMILES, a representation of molecular connectivity.

numSamples sets the number of structure candidates. The representative structure is the candidate with the highest aggregate_score; ties are resolved by the lowest candidate ID. This model score selects candidates and is not binding affinity or a probability of experimental success. pTM assesses confidence in the overall structural arrangement, while ipTM assesses interfaces between chains; the two metrics have different meanings.

There is a reason aggregate_score ranges from -100 to 1. The upstream definition is 0.2×pTM + 0.8×ipTM, which normally falls between 0 and 1, but 100 is subtracted when atoms of different chains are detected as clashing. A value near -99 therefore marks a clash rather than a gradual drop in confidence, and that candidate is effectively excluded from representative selection. The result table has no separate column for clashes, so read a negative score as that signal.

With a single-chain input there is no interface, so upstream ipTM is 0. The result table leaves ipTM blank, but aggregate_score is computed with that 0 in place, which brings the value down to 0.2×pTM. Do not read such a low aggregate score as a failed complex prediction.

Input settings

SettingFormatRequiredDefaultDescription and constraints
inputFormattaskOptional"sequence"Allowed values sequence · molecules · list
sequencesequenceConditionally required—Characters ACDEFGHIKLMNPQRSTVWYX:, Maximum length 2048, Applies to sequence
moleculesmoleculesConditionally required—JSON array, Applies to molecules · list
numSamplesnumberOptional1Number of structure candidates to generate. The candidate with the highest aggregate_score is selected as the representative structure. — Minimum 1, Maximum 64

Output metrics

ColumnDisplay nameRangeDescription
aggregate_scoreaggregate score-100~1From the upstream definition 0.2pTM + 0.8ipTM, 100 is subtracted when a clash between chains is detected. A value near -99 therefore marks a clash rather than low confidence
ptmpTM0~1—
iptmipTM0~1—

See Reading results for metric definitions and interpretation. A range shows possible values; whether higher or lower is better depends on the metric.

Execution availability

Open an execution plan in your project to see available tool versions and costs. Review and approve the inputs, settings, and terms before running.

License

Code, model weights, and accompanying data have their own terms. The table summarizes terms from official sources; check your execution plan for the configuration you will use. License texts are available in Third-party notices.

ComponentLicenseCommercial useSourceNotes
CodeApache-2.0Allowed, subject to notice requirementsLicense text—
Model weightsApache-2.0Allowed, subject to notice requirementsRepository READMEThe README states that both code and model weights are Apache 2.0 licensed
Model weightsApache-2.0Allowed, subject to notice requirementsWeight distribution metadataThe distributor’s model card lists the Chai-1 weights under the Apache 2.0 license.

Checked on 2026-10-06.

References

On this page