GIDetector

GIDetector detects genomic islands (GIs) in prokaryotic genomes to identify clusters of horizontally acquired genes associated with pathogenicity, antibiotic resistance, and environmental adaptation.

Key Features:

• Data Collection: Automates collection of prokaryotic genome data from public databases based on user specifications.
• Feature Analysis: Analyzes genomic features and integrates outputs from Alien_hunter and REPuter to detect anomalies such as atypical nucleotide composition and mobile genetic elements.
• Machine Learning Integration: Employs decision-tree-based ensemble classifiers that combine multiple GI-associated features to predict genomic islands.
• Ensemble Classifiers: Uses J48 decision trees as base classifiers and applies adaBoost, bagging, multiboost, and random forest ensemble algorithms to improve classification performance.
• Performance Evaluation: Evaluated on datasets from Salmonella, Staphylococcus, Streptococcus, and a mixed dataset, reporting that decision tree–based ensembles significantly improve GI classification across five performance evaluation metrics.

Scientific Applications:

• Bacterial Evolution: Enables investigation of bacterial evolution by identifying horizontally acquired genomic regions.
• Pathogenesis Studies: Supports analysis of pathogenicity by locating genomic islands associated with pathogenic traits.
• Antibiotic Resistance and Environmental Adaptation: Aids study of antibiotic resistance and environmental adaptation by identifying genomic islands implicated in these functions.
• Therapeutic and Environmental Intervention Development: Facilitates development of therapeutic strategies and environmental interventions by revealing GI-associated genetic mechanisms.

Methodology:

Collects prokaryotic genomes from public databases, extracts GI-associated features including outputs from Alien_hunter and REPuter (atypical nucleotide composition and mobile genetic element signals), and classifies candidate regions using decision-tree-based ensemble classifiers with J48 base classifiers and adaBoost, bagging, multiboost, and random forest.