Deep learning neural network development for the classification of bacteriocin sequences produced by lactic acid...

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Antibiotic resistance necessitates novel antimicrobial agents.
• Bacteriocins, particularly those from lactic acid bacteria (LAB), are promising natural alternatives.
• LAB bacteriocins are recognized as Generally Recognized As Safe (GRAS) and Qualified Presumption of Safety (QPS).

Purpose of the Study:

• To develop a deep learning model for classifying bacteriocins based on their LAB origin.
• To utilize interpretable k-mer features and embedding vectors for antimicrobial discovery applications.
• To enable the identification of novel bacteriocins with therapeutic potential.

Main Methods:

• A deep learning neural network was developed for binary classification of bacteriocin amino acid sequences.
• Features were extracted using k-mers (k=3,5,7,15,20) and vector embeddings (EV).
• The model was trained on Google Colab, demonstrating computational accessibility.

Main Results:

• The combination of 5-mers, 7-mers, and EV features achieved the highest performance (90.14% accuracy).
• Cross-validation demonstrated robust performance with low loss and high precision, recall, and F1 scores.
• Identification of conserved k-mer motifs specific to LAB bacteriocins was achieved.

Conclusions:

• The developed deep learning model shows high accuracy and outperforms existing methods.
• The model's feasibility in resource-limited settings is confirmed via cloud platforms.
• Identified k-mers can guide the design of synthetic antimicrobials, pending experimental validation.