Benchmarking Bioactive Peptide Predictors

From the long and short independent test sets, we extracted all PeptideDB peptide sequences. We refer to this set as the PeptideDB.70 dataset, as no peptide sequence in this dataset has 70% sequence similarity to any peptide in the training set. We divided these peptides into three activity subsets i.e. antimicrobial peptides, peptide hormones and toxin/venom peptides (see Table S12). We used the three activity subsets to compare the predictive power of PeptideRanker and two other state-of-the-art antimicrobial peptide predictors in each of these three bioactive peptide classes.