Caliby

1-5 minutesPaperCodeNeed >5GB

Potts model-based protein sequence design method that can condition on structural ensembles

Readiness

Readiness: Need >5GB
Smoke: passed
Artifacts: passed
Production: deferred_full_cuda_caliby_runtime
Storage: 5-20GB
Strategy: shared python312_slim.sif + tool ext3 overlay
Next: Build full GPU runtime, run real seq_des benchmark on bundled example CIFs, then wire real runner/API adapter and downloadable artifacts.

SOP agent staged upstream ProteinDesignLab/caliby revision 41d31560c3c73d7980d94f40f3c852b90bfab5c0 at /media/nik/seagate_nik/bio_server/tools/caliby/source. Source small (~56M) and model assets about 2.57GiB, but practical runtime needs large CUDA/PyTorch/Caliby env. Slurm job 374 failed missing pandas; job 377 saw shared torch CPU but missing pandas/lightning/hydra/omegaconf/biotite/atomworks/protpardelle. Evidence log: /media/nik/seagate_nik/bio_server/tools/caliby/evidence/caliby_sop_summary_20260701_224059/evidence_log.md.