RFantibody Walkthrough

EN / 中文

This page explains how the RFantibody module turns a structure-based antibody design request into ranked, downloadable candidates, and what remains before live GPU execution is production-ready.

Why this belongs in the platform
Right problem shape

RFantibody is useful when we know a target structure and want new VHH/scFv binders around a chosen epitope.

Clear handoff

The module produces structures, sequences, scores, and manifests that can flow into history, candidates, interface validation, and wet-lab review.

Compute isolated

The web/API contract is separated from GPU execution so the production runner can be Modal or a dedicated CUDA worker.

Operational flow
1
Upload target PDB and choose antigen hotspots.
target.pdb + hotspots
2
Pick a VHH/scFv framework preset or upload a custom framework.
framework preset / PDB
3
RFdiffusion-Ab generates antibody backbones around the hotspot patch.
backbone PDBs
4
ProteinMPNN samples CDR sequences on each backbone.
FASTA / sequence table
5
RF2-Ab predicts complexes and filters by pAE, RMSD, and optional ddG.
ranked CSV/JSON
6
Review top designs, download artifacts, then promote reviewed designs to Candidates.
candidate handoff
Platform mapping
Epitope Picker
Hotspot residue selection
RFantibody
Backbone and CDR design engine
Runs & Candidates
Source-tagged job history and ranked outputs
Interface
Downstream complex validation
Candidates
Reviewed design library
First-pass gates
RF2 pAE
< 10
Interface uncertainty gate
Self-consistency RMSD
< 2 A
Generated vs predicted complex consistency
Rosetta ddG
< -20
Optional binding-energy sanity check
Hotspots
2-8
Pilot range before broad campaigns
Runner and smoke test
What is ready now

The platform has a stable API contract, upload form, job table, artifact route, history integration, and runner-readiness banner.

Required smoke report

Before marking live execution complete, run a small target with 2-3 hotspots and 3 designs, then attach command, GPU image, weights, candidate count, top pAE/RMSD, artifacts, and failures.