Promera

New5-30 minutesPaperCodeNeed >5GB

Unified model for biomolecular cofolding (protein/RNA/DNA/ligand), binder filtering (iCS/ipSAE), and de novo binder design (minibinders + VHH nanobodies)

Readiness

Readiness: Need >5GB
Smoke: passed
Artifacts: passed
Production: deferred_large_cuda_weights
Storage: 5-20
Strategy: shared python312_slim.sif + tool ext3 overlay
Next: Handle in 5-20GB round with shared CUDA Torch 2.9/cuequivariance container, cached Hugging Face weights, tinyprot initialization, and a small schema inference benchmark.

SOP triage cloned upstream Promera revision 0840973a339029ac5c6ee9aef6970c6df2585b21 at /media/nik/seagate_nik/bio_server/tools/promera/source; source is 32M. Promera requires PROMERA_WEIGHTS from Hugging Face; checkpoint promera_2606.ckpt reports 1,885,729,053 bytes. pyproject requires Python >=3.12, torch==2.9.0, CUDA cuequivariance packages, tinyprot, and MSA initialization, so full validated install is expected to exceed the <5GB round even with shared containers.