Made by Brandon Thomason
Please paste your mature (without the signal peptide) heavy chain and/or light chain sequences to receive a ranked list of signal peptides scored against your construct.
How it works
The tool ranks candidates from a library of 11,019 mammalian signal peptides sourced from UniProt, each confirmed in vivo.
Each candidate is scored on three components. The model score comes from a Gradient Boosting model trained to predict SignalP D-score from 27 sequence features, including h-region hydrophobicity, n-region charge, and amino acid composition. The cleavage score measures how well the signal peptide's c-region matches the signal peptidase recognition site, scored specifically against the first two residues of your HC and/or LC mature sequence using a Position Weight Matrix built from 11,000 verified cleavage junctions. The h-region quality score penalises proline, which disrupts the hydrophobic helix, and cysteine, which can form disulfide bonds before the protein exits the translocon.
The composite score combines these three components with weights of 0.45 for model score, 0.45 for cleavage score, and 0.10 for h-region quality. Candidates are ranked by composite score by default, though you can re-sort by any individual component.
This tool narrows the search space to a shortlist of high-confidence candidates. Titer, product quality, and cell viability must be confirmed experimentally.