openfold3-structure-prediction

name	openfold3-structure-prediction
description	Run pinned OpenFold3 preview inference for proteins, nucleic acids, noncanonical residues, and small-molecule complexes. Use when an open AlphaFold3-style workflow is wanted and preview-status limitations are acceptable.
license	MIT

OpenFold3 Structure Prediction

Gate

Ask once before installing a pinned OpenFold3 preview release, running setup, downloading parameters/databases, contacting an MSA server, or using GPU/remote compute. Never send private sequences to public services without approval. Follow $cx-compute-environment.

Workflow

1. Define query JSON entities, sequences, ligand states, modifications, stoichiometry, templates, MSAs, and covalent relationships.
2. Pin the preview release/commit, parameter checksum, setup version, database snapshots, kernel versions, MSA/template mode, seeds, samples, and hardware.
3. Smoke-test the official example. Run run_openfold predict into artifacts/<run-id>/openfold3/; retain query JSON, feature inputs, structures, confidence outputs, environment, exact command, logs, and failures.
4. Check chemistry, atom mapping, clashes, chain/interface confidence, PAE-like uncertainty, sample diversity, and template dependence.
5. Record with $science-provenance; review claims with $science-review.

Boundaries

• Label the software and results as OpenFold3 preview; do not claim final-model parity or full AlphaFold3 modality parity.
• Predicted ligand poses and interfaces do not establish binding or affinity.