HyPe

HyPe predicts and classifies peptidoglycan hydrolases from complete open reading frames in genomic and metagenomic datasets to support discovery of potential antibacterial enzymes.

Key Features:

• Input scope: Processes complete open reading frames (ORFs) from genomic and metagenomic datasets.
• Feature representation: Uses amino acid composition and dipeptide composition features for classification.
• Machine learning models: Employs Random Forest and Support Vector Machines for training and optimization.
• Multiclass classification: Categorizes peptidoglycan hydrolases into multiple classes according to their site of action.
• Selected model performance: Random Forest multiclass model achieved up to 71.12% sensitivity, 99.98% specificity, 99.55% accuracy, and an MCC of 0.80 across four classes.
• Validation results: Validated on 24 independent genomic datasets with up to 100% sensitivity and an MCC of 0.94, and tested on 24 metagenomic datasets for novel hydrolase identification.
• Reported uniqueness: Described as the sole computational method for predicting peptidoglycan hydrolases from genomic and metagenomic data in the source description.

Scientific Applications:

• Hydrolase discovery: Identification and classification of novel peptidoglycan hydrolases from genomic and metagenomic ORFs.
• Antibacterial agent prioritization: Prioritization of candidate peptidoglycan hydrolases for experimental evaluation as potential antibacterial enzymes.
• Functional annotation: Functional annotation of ORFs with respect to peptidoglycan hydrolase classes and site of action.

Methodology:

Known peptidoglycan hydrolases were categorized by site of action; amino acid and dipeptide composition features were extracted and used to train Random Forest and Support Vector Machine classifiers, with a Random Forest multiclass model selected and validated on 24 genomic and 24 metagenomic datasets.