Boltz-2 (new)

New5-30 minutesPaperCodeProduction Ready

AlphaFold3 reproduction, predict protein/nucleic acid/small molecule complexes

Job Name

EXAMPLES

Use Boltz-2 ServerNumber of Samples * Predict affinity

Use YAML file inputNumber of RecyclesStep ScaleNumber of SeedsSeed (optional)Output FormatBoltz VersionMethodTemplate CIFsTemplate Distance Threshold Use Potentials Use Multiple Sequence AlignmentCustom A3M FilesMax MSA SequencesMSA Database Generate IPSAE metrics

All sequences must be non empty.

Readiness

Readiness: Production Ready
Smoke: passed
Artifacts: passed
Production: production_ready_real_boltz2_shared_sif_gpu_slurm_validated
Storage: 5-20GB shared runtime/cache installed
Strategy: Production Ready via shared python312_slim.sif, boltz_py312_site target, and shared Boltz cache under /media/nik/seagate_nik/bio_server/cache/boltz.
Next: Production Ready for protein-sequence Boltz-2 smoke/prediction path. Optional hardening: add UI controls for MSA server/custom MSA, ligand/CCD inputs, affinity property mode, and larger benchmark cases.

2026-07-18T17:33:47+08:00 SOP large-tool batch: promoted to Production Ready. Installed upstream boltz 2.2.1 with CUDA extras into /media/nik/seagate_nik/bio_server/envs/boltz_py312_site using shared python312_slim.sif. Shared runtime size 6.6GB; shared cache size 7.6GB with CCD data, Boltz-2 structure checkpoint, and Boltz-2 affinity checkpoint. Direct GPU benchmark on RTX 5050 passed for tiny_single_chain.yaml and produced CIF/confidence/PAE/PDE/pLDDT artifacts. Patched cluster_runner.py so runner_tool boltz no longer depends on missing /opt/containers/boltz.sif; it generates Boltz YAML from job_spec and runs shared SIF/site/cache. Runner API Slurm job b9eefdf16896 / Slurm 817 completed on gpu with artifacts summary.json, results.csv, prediction.cif, confidence.json, pae.npz, pde.npz, plddt.npz, boltz stdout/stderr, and empty slurm.err. prediction.cif 18,659 bytes; confidence_score 0.629489541053772; complex_plddt 0.7401653528213501.