AntiBMPNN
Type anything; we’ll convert to a safe ID when you run.
Upload PDB File
Drop your .pdb file here
Sequence Viewer
Design Regions
Short label for this AntiBMPNN job; shows in dashboard/history; no effect on results.
Input structure (.pdb). We read chains and residue numbering to enable region selection and design context. Clean, biologically relevant assemblies yield better designs.
Drop your .pdb file here
Interactive preview of detected chains/residues from the uploaded PDB. Use it to confirm numbering, select regions, and avoid off-by-one mistakes.
Choose chain(s) and residue ranges to redesign. Multiple rows allow multi-site design (CDR up to 6). Keep functional/structural residues fixed unless intentionally exploring.
Different checkpoints trained from different random seeds. That gives independently biased solutions; trying multiple weights is like an ensemble—useful for escaping local optima and boosting hit-rate diversity.
How many designs to sample. More samples increase the chance of high-quality sequences but cost more credits and time.
Controls diversity in amino-acid choices. ~0.1–0.3: very conservative/stability-oriented. 0.3–0.6: balanced default. 0.6–0.8: exploratory; expect more variation. 0.8–1.0: highly diverse/higher risk. 1.0: broad exploration—use sparingly. (ProteinMPNN-style sampling increases diversity as temperature rises.)
Adds small coordinate jitter before design to encourage robustness. 0.0: none. 0.05–0.15 Å: safe default. 0.15–0.30 Å: more exploration; can disrupt geometry. If designs overfit, raise noise slightly; if geometry degrades, lower it.