ARG-SHINE

ARG-SHINE classifies antibiotic resistance genes (ARGs) by integrating sequence homology, protein domain/family/motif annotations, and raw amino acid sequences with a deep convolutional neural network and a learning-to-rank ensemble to improve detection of ARGs divergent from reference databases.

Key Features:

• Integration of Diverse Data Sources: Combines sequence homology, protein domain/family/motif functional information, and raw amino acid sequences for comprehensive analysis.
• Deep Convolutional Neural Network (CNN): Applies a deep CNN to raw amino acid sequences to capture complex sequence patterns relevant to ARG function.
• Learning-to-Rank Ensemble: Uses a learning-to-rank machine-learning approach to ensemble three distinct component methods and produce a ranked set of predictions.
• Component Method Diversity: Ensembles three component methods that each focus on different feature sets to enhance robustness and complementarity.
• Robustness to Divergent ARGs: Designed to improve classification performance for ARGs that diverge from sequences present in existing reference databases.
• Performance Metrics: Reported improvements include superior accuracy, macro-average F1-score, and weighted-average F1-score on benchmark datasets.

Scientific Applications:

• Enhanced ARG Classification: Improves identification and classification of ARGs to aid studies of microbial community dynamics and inform treatment strategies for bacterial infections.
• Functional Screening Support: Supports classification of newly discovered ARGs from functional screening, maintaining prediction accuracy when genes are underrepresented in reference databases.

Methodology:

ARG-SHINE extracts and integrates sequence homology and functional annotations (protein domains/families/motifs) with raw amino acid sequences, analyzes raw sequences using a deep convolutional neural network, and combines outputs of three component methods via a learning-to-rank ensemble.