Deep-ABPpred: identifying antibacterial peptides in protein sequences using bidirectional LSTM with word2vec

Area of Science:

• Biotechnology
• Bioinformatics
• Computational Biology

Background:

• Antibiotic overuse has driven the emergence of antimicrobial resistance, necessitating alternative therapeutic strategies.
• Antibacterial peptides (ABPs) are a promising alternative, but their identification from natural sources is laborious and expensive.
• Developing in silico models is crucial for efficient identification of novel ABPs for synthesis and testing.

Purpose of the Study:

• To propose Deep-ABPpred, a deep learning classifier for identifying antibacterial peptides (ABPs) in protein sequences.
• To evaluate Deep-ABPpred's performance against existing state-of-the-art ABP classifiers.
• To demonstrate the practical application of Deep-ABPpred in discovering and validating novel ABPs.

Main Methods:

• Development of Deep-ABPpred using a bidirectional long short-term memory (LSTM) algorithm.
• Utilizing amino acid level features derived from word2vec embeddings for model training.
• Validation on both test and independent datasets to assess predictive accuracy.

Main Results:

• Deep-ABPpred achieved high precision, approximately 97% on the test dataset and 94% on the independent dataset.
• The model outperformed other state-of-the-art ABP classifiers in identifying antibacterial peptides.
• Identified ABPs from Streptococcus bacteriophage tail proteins demonstrated potent in vitro antibacterial activity.

Conclusions:

• Deep-ABPpred is a highly accurate and efficient tool for in silico identification of novel antibacterial peptides.
• The model's predictive capability was validated through the successful synthesis and testing of newly identified ABPs.
• An accessible online prediction server (https://abppred.anvil.app/) has been developed to facilitate the use of Deep-ABPpred.