ParaSurf

5-30 minutesPaperCodeProduction Ready

Predict paratope residues

Readiness

Readiness: Production Ready
Smoke: passed
Artifacts: passed
Production: production_validated_runner_api_slurm_parasurf
Storage: <5
Strategy: compiled/binary SIF or shared python312_slim + ext3 overlay
Next: Production Ready. Optional hardening: cache/upload smaller validated antibody inputs for faster CI, expose mesh density/device controls in UI, and suppress OpenBabel warning noise.

Google Drive model weight download via gdown did not complete. 2026-07-09T13:11:09+08:00 SOP production validation: promoted from Production Blocked to Production Ready. Built a real ParaSurf adapter using /media/nik/seagate_nik/bio_server/tools/parasurf/ParaSurf, shared python312_slim.sif, palm_py312_site, itsflexible OpenBabel runtime with compat libraries, a small /media/nik/seagate_nik/bio_server/tools/parasurf/py312_site target for jsonpickle/torchsummary/h5py, PECAN weights /media/nik/seagate_nik/bio_server/weights/parasurf/pecan_best_weights.pt, and bound DMS runtime /media/nik/seagate_nik/bio_server/tools/parasurf/dms_runtime at /usr/local/lib/dms. Validated upstream blind_predict.py on bundled 4N0Y_receptor_1.pdb through Ubuntu runner API job 3aaecb38fd37 / Slurm job 586. Run completed in 999.562 s and produced residue and per-atom prediction PDBs with 3,271 finite scores each, score range 0.001-0.988, plus one extracted pocket PDB with 132 atoms. Artifacts verified: summary.json, results.csv, parasurf_pockets.csv, parasurf_result.json, parasurf_residue_prediction.pdb, parasurf_atom_prediction.pdb, parasurf_pocket1.pdb, parasurf_centers.txt, parasurf_stdout.txt, parasurf_stderr.txt, slurm.out, slurm.err.