StaBle-ABPpred

StaBle-ABPpred predicts antibacterial peptides (ABPs) from protein sequences using a stacked ensemble of deep and classical machine learning models to accelerate discovery of candidates against multi-drug resistant (MDR) bacteria.

Key Features:

• Stacked ensemble architecture: Integrates deep learning and traditional machine learning models in a multi-layer ensemble to improve prediction robustness.
• Base-level biLSTM with attention: Uses bidirectional long short-term memory networks combined with an attention mechanism to capture sequence patterns in peptides.
• Meta-level ensemble classifiers: Aggregates outputs using random forest, gradient boosting, and logistic regression classifiers.
• Performance evaluation: Compared against state-of-the-art classifiers with statistical testing using analysis of variance (ANOVA) and post hoc analyses reporting improved accuracy and precision.
• Sequence-level peptide identification: Identifies novel ABPs and detects amino acid similarities to experimentally validated antimicrobial peptides.

Scientific Applications:

• Genome-scale ABP discovery: Screening of genomes and proteomes to identify candidate antibacterial peptides, exemplified by analysis of the Streptococcus phage T12 genome.
• Candidate prioritization for validation: Prioritizing peptides for chemical synthesis and experimental testing against MDR bacterial strains.
• Exploration of protein diversity: Mining diverse proteins to uncover broad-spectrum antibacterial peptide candidates.

Methodology:

Stacked ensemble combining base-level bidirectional LSTM networks with an attention mechanism and a meta-level ensemble of random forest, gradient boosting, and logistic regression classifiers, with performance assessed by comparison to state-of-the-art models and statistical evaluation via ANOVA and post hoc analyses.