RFD4-Proteina

Overview

Protein binder design against a target structure with backbone geometry, sequence, and hotspot conditioning.
Motif scaffolding from an input structure using a contig specification.
De novo monomer generation from a length range.
One foundation model handles monomers, binders, antibodies, enzymes, DNA binders, and small-molecule pockets.

Modes

Mode	Input shape	When to use it
`binder_design` Binder Design Default	Uses one selected file/source when file parameters are present.	Design binders against a target structure using backbone geometry, sequence, and hotspot conditioning.
`motif_scaffolding` Motif Scaffolding	Uses one selected file/source when file parameters are present.	Scaffold a preserved motif from an input structure using a contig specification.
`unconditional_monomer` Unconditional Monomer	No uploaded input is required by the mode itself.	Generate de novo monomers from a requested length range.

Canonical Job Configuration

These are the fields exposed by the default job configuration for rfd4-proteina. They are also returned by GET /api/v1/program/params?program=rfd4-proteina and submitted as the params JSON object to POST /api/v1/job/submit.

Parameter	Type	Modes	What it does
`target_structure` Target Structure	Structure file	Binder Design	Choose the target structure from an upload, dataset, or previous job output. RFD4-Proteina reads CIF files. Required; Files: .cif
`binder_length` Binder Length	Length or range	Binder Design	Length or length range for the generated binder chain. Required; Range: 3-500
`hotspots` Hotspots	Residue selection	Binder Design	Optional target residues the binder should contact.
`motif_structure` Motif Structure	Structure file	Motif Scaffolding	Structure holding the motif to scaffold. RFD4-Proteina reads CIF files. Required; Files: .cif
`contig` Contig	Contig string	Motif Scaffolding	Which residues of the input to keep and where to insert expandable segments, e.g. A1-50, LINK:10-30. Required
`motif_fix` Fix Motif	Text	Motif Scaffolding	Condition on all motif atoms or only the protein backbone. Default: all_atom; Options: all_atom, backbone_only
`length` Length	Length or range	Unconditional Monomer	Length or length range for the generated monomer. Required; Range: 20-500
`num_designs` Designs	Integer	All modes	How many designs to generate. Default: 10; Range: 1-100
`recycle_strategy` Thinking Budget	Text	All modes	Trunk recycling schedule. Higher budgets cost more time but can improve design quality. Options: thinking_low, thinking_medium, thinking_high

Outputs And Metrics

Generated all-atom designs as CIF files with standard RFD4 annotations (motif provenance, conditioning).
Canonical structure copies are published for pipelines and downstream structure consumers.
RFD4-Proteina outputs are generated candidates. Use downstream prediction, interface confidence, clash checks, and filtering to prioritize.

Common Examples

Binder design: target CIF, binder length 50-120, hotspots A64,A88, 10 designs.
Motif scaffolding: motif CIF, contig A1-50, LINK:10-30, backbone-only fixing.
Unconditional monomer: length 100-150, 10 designs.

Example API params

{
  "mode": "binder_design",
  "target_structure": "target.cif",
  "binder_length": "50-120",
  "hotspots": "A64,A88",
  "target_segments": "A17-131",
  "num_designs": 10
}

Caveats

Structure inputs must be CIF files; convert PDB targets before submitting.
Hotspot and target-segment selections must match the uploaded target numbering.
Inline metrics and refolding scores are not wired up yet; filter designs downstream.
Full precision is used by default so the same configuration runs on every GPU pool; mixed precision can be enabled on newer GPUs through advanced submit.

Advanced Submit

Advanced submit is still available for direct program arguments through POST /api/v1/job/submit-advanced. Prefer canonical configuration unless you need exact low-level arguments or are reproducing a known command line.

Advanced submit exposes annotated CIF and condition-spec inputs plus sampling overrides for specialized designs.
Use num_replicates for design count when working at the direct-argument level.

curl -X POST https://subseq.bio/api/v1/job/submit-advanced \
  -H "Authorization: Bearer <api_key>" \
  -F program=rfd4-proteina \
  -F 'args=inputs=/inputs/design.yaml'