CVAE training data and generated peptide/enzyme candidates for AMP-CPP hybrid peptide design

Training datasets, configuration, and generated candidates for a calibrated dual-score generative pipeline (CDSPO) designing hybrid antimicrobial/cell-penetrating peptides. This dataset accompanies the CDSPO (Calibrated Dual-Score Peptide Optimisation) pipeline for de novo design of peptides with combined antimicrobial (AMP) and cell-penetrating (CPP) activity. Contents include: curated AMP positive training data (n = 1,655, ADP v6 disulfide-bond entries), curated CPP positive training data (n = 709, CellPPD/Raghavendra collection), YAML configuration file specifying all model hyperparameters, training settings, and filter thresholds used to produce the reported results, final set of 80 generated hybrid AMP/CPP peptide candidates with predicted AMP/CPP probabilities and physicochemical properties. Candidates were produced via a conditional variational autoencoder coupled to calibrated classifier ensembles (gradient boosting, multilayer perceptron, random forest) and score-guided latent optimisation.