HalluDesign
Design protein and biomolecular complexes with HalluDesign.
Results
Loading starting example…
Method & evidence
Run the released Protenix design matrix from starting structures or random initialization with protein, ligand, DNA, RNA, symmetry, cyclic, framework, and sequence-bias controls.
Scope & limitations
- AF3 and cross-model paths are excluded from this release; the released runtime is Protenix-only.
- CCD conditioning that is ignored by the Protenix no-PDB builder is rejected rather than silently accepted.
Interpreting results
- Generated complexes are computational hypotheses and do not establish binding, expression, stability, or function.
- Compare outputs only under identical conditioning, sequence-model, recycling, and sampling settings.
- Inspect ligand or nucleic-acid placement and protein geometry before downstream selection.
References
Input requirements & outputs
Input checks
- One to eight checksum-bound homogeneous rows
- Supported conditioners are explicit SMILES, DNA, RNA, fixed residues/chains, bias, framework/CDR, symmetry, cyclic, and redesign controls
- At most 32 random-init trajectories or eight starting-structure rows per native batch
Expected results
- Designed sequence and complex table — Review row identity, design identity, sequence, status, and model evidence.
- All-atom structures — Inspect every successful Protenix PDB result in the interactive structure viewer.
- Conditioning and provenance evidence — Retain exact templates, constraints, image, assets, request, result, and artifact digests.
