DefPred

DefPred predicts defensins in protein sequences using SVM-based machine learning models to distinguish defensins from antimicrobial peptides and non-defensin proteins for applications in innate immunity and antimicrobial peptide research.

Key Features:

• Main dataset: A primary dataset comprising 1,036 defensins and 1,035 antimicrobial peptides (AMPs) was used for model development.
• Alternate dataset: A Swiss-Prot–derived dataset comprising 1,036 defensins and 1,054 non-defensins was used as an alternate training and validation set.
• Residue analysis: Identification of key residues—Cys, Arg, and Tyr—that are more abundant in defensins compared to AMPs.
• Peptide features: Models were built using a range of peptide features (feature types not further specified in the input).
• Machine learning models: Support vector machine (SVM)-based models were developed to classify defensins versus non-defensins/AMPs.
• Performance on main dataset: Validation performance reached an MCC of 0.88 and an AUC of 0.98.
• Performance on Swiss-Prot dataset: Validation performance reached a maximum MCC of 0.96 and an AUC of 0.99.
• Functional capabilities: Predicting presence of defensins in protein sequences, scanning proteins for defensin regions, and designing optimal defensins from analogs.

Scientific Applications:

• Defensin identification: Detecting and classifying defensin sequences within protein datasets.
• Protein scanning: Locating defensin regions within larger protein sequences.
• Defensin design: Guiding design or optimization of defensin analogs.
• Innate immunity research: Studying host defense peptides and their sequence determinants.
• Antimicrobial peptide and antibiotic resistance studies: Supporting research into AMPs and strategies addressing antibiotic resistance.

Methodology:

A systematic machine learning approach used peptide feature extraction, residue enrichment analysis (Cys, Arg, Tyr), and SVM-based classification trained and validated on a main dataset (1,036 defensins, 1,035 AMPs) and a Swiss-Prot alternate dataset (1,036 defensins, 1,054 non-defensins) with performance evaluated by MCC and AUC.