rfdiffusion-protein-design

name	rfdiffusion-protein-design
description	Run pinned RFdiffusion for unconditional generation, motif scaffolding, binder backbones, symmetric assemblies, partial diffusion, or macrocyclic peptide design. Use when structural conditioning and a downstream sequence/refolding validation funnel are defined.
license	MIT

RFdiffusion Protein Design

Use $cx-modeling-problem-execution for a concrete target or motif and continue through the complete candidate funnel.

Gate and workflow

1. Ask once before cloning a pinned commit/container, downloading checkpoints, and running GPU compute. State candidate count, expected time/storage, and any downstream ProteinMPNN/refolding compute.
2. Define design task, target/motif provenance, hotspots, contig map and indexing, fixed residues/chains, length/topology/symmetry, potentials, diffusion steps, seeds, and candidate budget. Pin checkpoint and config.
3. Smoke-test the matching official example and a tiny pilot. Run backbones into artifacts/<run-id>/rfdiffusion/; preserve PDB/TRB trajectories, mappings, configs, logs, rejects, and all candidate IDs.
4. Chain to $cx-proteinmpnn-sequence-design, independent refolding/complex prediction, geometry/interface filters, diversity/novelty checks, and developability screens. Report the full funnel and orthogonal-model agreement.
5. Record with $science-provenance; review with $science-review before ranking.

Boundaries

• A generated backbone or computational pass does not establish folding, function, binding, specificity, safety, or experimental success.
• Never score only survivors; report generation and filter failure rates.