FrogPCSP: a propeptide cleavage site predictor for frog antimicrobial peptides

Frogs produce and secrete cutaneous antimicrobial peptides (AMPs), which serve as chemical defense against microbial infections and have great biotechnological potential. AMPs are stored in intracellular vesicles in skin glands as propeptides. Before secretion, proprotein convertases (PCs) cleave propeptides into acidic spacer peptides and bioactive peptides. Identifying the correct cleavage site between the acidic spacer and bioactive peptides is a crucial step in AMP prediction. Here, we present Frog Propeptide Cleavage Site Predictor (FrogPCSP), an SVM-based predictor designed to identify propeptide cleavage sites in frog AMPs. The SVM model showed strong performance (global accuracy = 0.981, precision = 0.937, recall = 0.928, F1-score = 0.933, PR-AUC: 0.916) under grouped and stratified 10-fold cross-validation on 424 positive and 2488 negative cleavage sites. Overall, FrogPCSP demonstrated superior performance for propeptide cleavage site prediction of frog AMPs (AUC = 0.990) relative to PSSM (AUC = 0.974) and ProP (AUC = 0.905), a general-purpose reference prohormone cleavage site predictor. As proof of concept, 595 unlabeled frog AMP sequences from UniProtKB/TrEMBL were analyzed. FrogPCSP inferred 926 putative propeptide cleavage sites. Computational physicochemical profiling of the resulting peptides revealed two distinct clusters, one positively charged, with high isoelectric point and strong amphipathicity-consistent with putative bioactive peptides-and another negatively charged, with low isoelectric point and low amphipathicity values-corresponding to putative acidic spacer peptides. Thus, we believe identifying propeptide cleavage sites will assist the discovery and advance the understanding of novel frog AMPs.