BaPreS

BaPreS predicts bacteriocin protein sequences using machine learning and an optimized set of sequence-derived features to identify novel ribosomally synthesized antimicrobial peptides.

Key Features:

• Feature Extraction: Extracts sequence-derived physicochemical and structural features including amino acid composition and hydrophobicity from known bacteriocin and non-bacteriocin protein sequences.
• Feature Optimization: Applies statistical justification and recursive feature elimination to select an optimal subset of informative features.
• Machine Learning Models: Trains Support Vector Machine (SVM) and Random Forest (RF) classifiers on the selected feature set to discriminate bacteriocin versus non-bacteriocin sequences.
• Performance Evaluation: Evaluates model performance on an established dataset, reporting 95.54% prediction accuracy on testing protein sequences.
• Comparative Performance: Demonstrates higher predictive performance than traditional sequence matching-based tools and contemporary deep learning methods.
• Training Data Augmentation: Supports expansion of the training dataset by adding bacteriocin and non-bacteriocin sequences to update and improve model performance.

Scientific Applications:

• Bacteriocin Discovery: Identifies candidate bacteriocin sequences for experimental validation and characterization.
• Antibiotic Development: Supports the selection of novel ribosomally synthesized antimicrobial peptides for development against antibiotic-resistant bacteria.
• Microbiology and Biotechnology Research: Enables comparative analyses of peptide features and aids bioinformatics studies in microbial peptide repertoires.

Methodology:

Extracts physicochemical and structural features (including amino acid composition and hydrophobicity) from labeled bacteriocin and non-bacteriocin sequences, performs statistical justification and recursive feature elimination for feature selection, trains SVM and Random Forest classifiers on the selected features, and evaluates performance on an established testing dataset reporting 95.54% accuracy.