BaPreS: a software tool for predicting bacteriocins using an optimal set of features.

Area of Science:

• Biochemistry
• Computational Biology
• Microbiology

Background:

• Antibiotic resistance is a global health crisis necessitating novel antimicrobial agents.
• Bacteriocins show promise as antibiotics but are challenging to identify due to sequence diversity.
• Traditional sequence matching methods struggle to detect novel, dissimilar bacteriocins.

Purpose of the Study:

• To develop a machine learning-based software tool, BaPreS (Bacteriocin Prediction Software), for accurate bacteriocin protein sequence detection.
• To identify an optimal set of features for enhanced bacteriocin prediction.
• To create a user-friendly tool for discovering new bacteriocins.

Main Methods:

• Extracted physicochemical and structural protein features from known bacteriocin and non-bacteriocin sequences.
• Reduced feature sets using statistical methods and recursive feature elimination.
• Developed and compared Support Vector Machine (SVM) and Random Forest (RF) models, selecting the best for implementation.

Main Results:

• The BaPreS software achieved a prediction accuracy of 95.54% on testing protein sequences.
• BaPreS demonstrated superior performance compared to sequence matching and deep learning methods.
• The tool allows for user-added sequences to improve future predictive accuracy.

Conclusions:

• BaPreS is an effective tool for discovering novel, dissimilar bacteriocins, crucial for developing new antibiotics.
• The software is compatible with Windows, Linux, and macOS operating systems.
• An open-source package and user manual are available for public access and use.