Deep-ABPpred

Deep-ABPpred predicts antibacterial peptides (ABPs) within protein sequences using deep learning to accelerate discovery of candidate antimicrobial peptides.

Key Features:

• Deep Learning Classifier: A bidirectional long short-term memory (BiLSTM) classifier captures sequential dependencies in protein sequences.
• Amino Acid Level Features: Amino-acid-level features are encoded using word2vec embeddings to represent sequence context.
• High Precision Performance: Reported precision is approximately 97% on test datasets and 94% on independent datasets.
• Experimental Validation from Predictions: Predicted peptides from Streptococcus bacteriophage tail proteins were chemically synthesized and exhibited in vitro antibacterial activity against Gram-positive and Gram-negative bacteria.

Scientific Applications:

• ABP Discovery: Prioritizing candidate antibacterial peptides from protein sequence data for downstream validation.
• Phage-derived Peptide Screening: Screening Streptococcus bacteriophage tail proteins to identify potential ABPs.
• Experimental Prioritization: Selecting peptides for chemical synthesis and in vitro testing to find agents active against Gram-positive and Gram-negative bacteria.

Methodology:

A bidirectional LSTM (BiLSTM) deep learning classifier trained on amino-acid-level word2vec embeddings that produces probability scores for ABP predictions.