DeepBL

DeepBL classifies and predicts beta-lactamase (BL) enzymes from protein and genome sequence data to identify antibiotic-resistance determinants.

Key Features:

• Deep learning architecture: Implements a Small VGGNet-based deep neural network using the TensorFlow library.
• Input data: Processes protein and genome sequence data as model inputs.
• Sequence-derived features: Incorporates sequence-derived features to enhance predictive accuracy.
• Multi-class classification: Performs multi-class classification to distinguish different classes of beta-lactamases.
• High-throughput processing: Designed for large-scale, high-throughput classification and prediction workflows.
• Evaluation framework: Performance assessed using 10-fold cross-validation and independent test datasets on curated benchmark datasets.
• Sequence redundancy tuning: Enables experiments with varying sequence redundancy thresholds in training datasets.
• Negative sample selection: Employs strategic negative sample selection in benchmark dataset construction.
• Proteome-wide screening: Applied to proteome-wide screening of reviewed bacterial protein sequences from UniProt.
• Generalization: Demonstrates strong generalization capabilities across different datasets.

Scientific Applications:

• Beta-lactamase identification: Identification and classification of beta-lactamase enzymes from sequence data.
• Antibiotic-resistance research: Investigation of beta-lactam antibiotic resistance mechanisms in bacterial pathogens.
• Proteome-wide surveys: Large-scale screening of UniProt reviewed bacterial proteins to detect BL enzymes.
• Experimental prioritization: Prioritization of candidate BLs for experimental validation.
• Benchmarking: Benchmarking and evaluation of predictive models for BL classification.

Methodology:

Multi-class classification using a Small VGGNet architecture implemented in TensorFlow, incorporating sequence-derived features, evaluated by 10-fold cross-validation and independent test datasets on curated benchmarks, with experiments on sequence redundancy thresholds and strategic negative sample selection.