Caliby
Design and score structure-conditioned protein sequences.
Results
Loading starting example…
Method & evidence
Design or score sequences against one structure or an aligned conformational ensemble with the five released Caliby model profiles.
Scope & limitations
- All four released operations and all five model profiles were exercised on the qualified H100 image.
- Designed sequences require independent structure prediction and experimental validation.
Interpreting results
- Lower Caliby energy means greater compatibility under the selected learned Potts model; it is not an experimental stability or affinity measurement.
- Compare scores only when operation, model profile, input structures, and sampling settings are the same.
- Ensemble results require residue-aligned conformers; misaligned structures are rejected.
References
Input requirements & outputs
Input checks
- PDB coordinates with a consistent residue and chain mapping
- Scoring requires one canonical amino-acid sequence matching the structure length
- Design requests may generate 1–64 sequences with an explicit seed
Expected results
- Designed sequences and scores — Inspect every generated or submitted sequence with whole-protein Caliby energy.
- Residue-level energy table — Review per-position energy evidence in the complete result Artifact.
- Native request and model evidence — Retain operation, model checkpoint, inputs, parameters, and result digests.
