ProteinMPNN

5-30 minutesPaperCodeProduction Ready

Protein structure to sequence prediction

Readiness

Readiness: Production Ready
Smoke: passed
Artifacts: passed
Production: production_pass
Storage: <5
Strategy: shared PyTorch SIF + tool ext3 overlay; avoid per-tool torch duplication
Next: Production Ready. Optional next hardening: add richer UI controls for chain/fixed-position design and GPU acceleration for larger batches.

Batch A evidence: upstream ProteinMPNN source at /media/nik/seagate_nik/bio_server/tools/proteinmpnn/repo, repo/tool folder 228MB. Shared PyTorch CPU package environment is /media/nik/seagate_nik/bio_server/envs/sequence_design_torch_cpu_py312, about 6.1GB, used with shared python312_slim.sif. Slurm jobs 259 and 260 failed from wrapper/container path issues; Slurm job 261 passed using upstream parse_multiple_chains.py plus protein_mpnn_run.py on bundled 5L33 and 6MRR monomer PDBs. Generated valid seqs/5L33.fa and seqs/6MRR.fa with sample scores 0.8576 and 0.9498. Backend validation passed, but website runner adapter is not wired yet. Production evidence 2026-07-01: real runner adapter wired to runner_tool proteinmpnn and submitted through cluster runner API to Slurm job 265. Benchmark used bundled upstream PDB monomers 5L33 and 6MRR via shared Python 3.12 SIF plus shared PyTorch CPU site-packages. Runtime_seconds 3.726. Artifacts: summary.json, results.csv, designed_5L33.fa, designed_6MRR.fa, proteinmpnn stdout/stderr, copied PDB inputs. Scientific sanity: 2 targets, 2 designed sequences, best ProteinMPNN score 0.7940; FASTA artifacts nonempty and downloadable through runner artifact manifest.