Proteina-Complexa

De novo binder design against protein or small-molecule targets, run as an inference-time search filtered by an AF2 / RF3 / force-field reward stack. Fans out as a fund-and-drain campaign of independent search shards; the wallet balance is the only ceiling.

What it is for

Pick Proteina-Complexa when you want de novo binders against a protein OR a small-molecule (ligand) target, scored by a full AF2 / RF3 / force-field reward stack, and you want to scale the search across many GPUs with the wallet as the only ceiling.

Proteina-Complexa (Geffner et al., NVIDIA 2025). A flow-matching generator wrapped in an inference-time search that filters designs through an AlphaFold2 / RoseTTAFold3 / force-field reward stack. It designs de novo binders against protein targets, small-molecule (ligand) targets, and enzyme / motif active sites. Runs here as a fund-and-drain campaign of independent seeded search shards, with global cross-shard top-K and post-hoc diversity clustering.

When it fits:

You want to aim a de novo binder at a specific epitope, including a recessed or occluded one, using hotspot residues.
Your target is multi-chain and the site you care about spans more than one chain.
You want de novo binders against a small-molecule target, not just a protein.
You want an inference-time search filtered by an AF2 / RF3 / force-field reward, not raw generation.
You want to scale the search across many GPUs with the prepaid wallet as the only ceiling.
You want diverse high-reward designs (global top-K + diversity clustering) rather than near-duplicates.

Inputs

You will need:

A target: your own structure (.pdb/.cif) for the protein-binder variant, or a curated benchmark task for any variant.
For your own target, the chain ID — or a chain/residue range such as A1-150, or A12-157,B12-157,C12-157 for a multi-chain target.
Optionally, hotspot residues to aim the binder at a specific epitope.
A funded wallet that covers at least the first wave of shards.

Each run uses a preset that sets the scale and scope:

Protein binder (de novo, vs a protein target): Design de novo binders against a protein target. Search is scored by AlphaFold2 confidence plus a force-field reward. Pick a curated target task or upload your own target PDB.
Ligand binder (de novo, vs a small molecule): Design de novo binders against a small-molecule target supplied as an SDF. Scored by the RoseTTAFold3 reward (the force field does not support protein-ligand complexes). Pick a curated ligand task or upload your own SDF.
Motif scaffolding / enzyme (AME): Scaffold a functional motif or enzyme active site. Scored by the RoseTTAFold3 reward. Pick a curated AME task or upload your own motif.
Validate (free dry-run): Free CPU-only pre-flight that checks your target + config load before you commit GPU to a paid search. No wallet charge.

Parameters you set on the form:

Design variant: protein_binder for a protein target (AF2 reward), ligand_binder for a small-molecule SDF target (RF3 reward), motif_ame for motif scaffolding / enzyme active sites, or validate for a free config check before spending GPU.
Target: Your own structure, or a curated benchmark task whose target is baked in — the two are mutually exclusive. Uploading your own is available on the protein-binder variant; the ligand and motif variants run curated tasks (their tasks resolve from separate upstream registries).
Target region: Which chains and residues to design against, e.g. A1-150, or A12-157,B12-157,C12-157 for a multi-chain target. Blank uses the whole target chain.
Hotspot residues: Optional. Residues the binder should contact, in original PDB numbering — plain numbers use the target chain, or prefix the chain (A113 C73) for a multi-chain region. Every hotspot is checked against your structure before any GPU runs, so a residue that is not there is refused rather than quietly ignored.
Binder length: The range each design's length is drawn from. Defaults to 60-120 residues.
Number of designs: How many designs to search for. This scales the number of independent search shards; each shard runs on its own GPU and returns its survivors, and the hub picks the global top set.

Typical runtime:

protein_binder: ~9 to 15 min / shard (measured at 130-415 residues)
ligand_binder: not yet measured (under 120 min / shard)
motif_ame: not yet measured (under 120 min / shard)
validate: 1 to 3 min (free)

How to read the results

Ranked designs with reward scores (AF2 pLDDT / ipTM for protein, RF3 score for ligand / motif, force-field energy where applicable), a structural diversity cluster id, and downloadable structures. The ligand and motif variants score on RF3 only.

Where a tool reports them, the scores mean:

ipTM: Predicted confidence at an interface. Higher is better; aim above roughly 0.7 on a tractable target. On a multi-chain target the number may cover the target's own chain–chain interface as well as the binder's, so read the per-tool note on the results table before comparing designs on it.
pLDDT: Per-residue confidence in the predicted fold. Higher means the model is more sure of that part of the structure.
i_pAE and pAE: Predicted alignment error, at the interface (i_pAE) or across the whole structure (pAE). Lower is better.

References

Geffner et al., NVIDIA (2025)

Read the paper Source code on GitHub

Open the Proteina-Complexa form All guides