PPAP

1-5 minutesPaperCodeNeed >5GB

Protein-protein binding affinity prediction for multiple complexes

Readiness

Readiness: Need >5GB
Smoke: passed
Artifacts: passed
Production: deferred_esm2_3b_weights_cuda_graph_env
Storage: 5-20GB
Strategy: compiled/binary SIF or shared python312_slim + ext3 overlay
Next: Build or reuse a shared PyTorch/PyG/ESM container, mirror the PPAP checkpoint and ESM2-3B assets, patch CPU/GPU device selection, run example PDBs through cal_graph.py and PPAP_test.py, then add a real runner adapter and artifact validation.

SOP triage cloned upstream PPAP at /media/nik/seagate_nik/bio_server/tools/ppap/source, revision a1e9bc5a3ba85f348ac8c73983e904bb3c78a7ee; source is about 23M. Production path requires ESM2-3B embeddings, PPAP checkpoint from Hugging Face qj666/PPAP, Python3.10/PyTorch/PyG/Lightning stack, graph feature generation, and CUDA-hardcoded PPAP_test.py device handling. No lightweight benchmark run.