ThermoMPNN

1-5 minutesPaperCodeProduction Ready

Protein mutation recommendation for increased thermostability

Readiness

Readiness: Production Ready
Smoke: passed
Artifacts: passed
Production: production_pass
Storage: <5
Strategy: shared python312_slim.sif + ThermoMPNN site_cpu target packages and staged checkpoints
Next: Production Ready. Optional hardening: expose PDB upload and chain selector in UI, add output filtering for top stabilizing/destabilizing mutations, and add GPU path for larger batches.

Batch D evidence: source cloned at /media/nik/seagate_nik/bio_server/tools/thermompnn/source, storage 245MB. Upstream benchmark is analysis/custom_inference.py --pdb examples/2OCJ.pdb --chain A --model_path models/thermoMPNN_default.pt. Tool-local venv setup failed because the Seagate filesystem rejects virtualenv symlinks; pip --target install was interrupted. No Slurm production job has passed yet. 2026-07-07 16:05 +0800: SOP Runner Adapter Pending audit: demoted from Runner Adapter Pending because only placeholder thermompnn-smoke is configured and the official 2OCJ ThermoMPNN benchmark has not passed through Slurm; runtime setup/checkpoint staging remain incomplete. 2026-07-08T22:53:32+08:00 SOP production validation: promoted from Production Blocked to Production Ready. Staged ThermoMPNN dependencies into /media/nik/seagate_nik/bio_server/tools/thermompnn/site_cpu and patched local.yaml to use the shared backend source path. Validated upstream Kuhlman-Lab/ThermoMPNN analysis/custom_inference.py on official examples/2OCJ.pdb chain A with models/thermoMPNN_default.pt through Ubuntu runner API job 9bb77cabc966 / Slurm job 572 using shared python312_slim.sif. Run completed in 8.022 s and produced 3,880 finite site-saturation ddG predictions; min ddG -1.8490122556686401 for A63I, max ddG 4.759126663208008 for I159D. Artifacts verified: summary.json, results.csv, ThermoMPNN_inference.csv, thermompnn_stdout.txt, thermompnn_stderr.txt, 2OCJ.pdb, slurm.out, slurm.err.