Classification of genomic islands using decision trees and their ensemble algorithms.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Genomic islands (GIs) are distinct clusters of alien genes found in some bacterial strains but absent in others of the same genus.
• Accurate identification of GIs is vital for medical and environmental applications, yet current detection methods remain insufficient.
• Understanding GI distribution aids in tracking bacterial evolution and pathogenicity.

Purpose of the Study:

• To evaluate and compare the effectiveness of various machine learning approaches for classifying genomic islands (GIs).
• To assess the performance of decision tree algorithms and their ensemble methods on GI datasets from Salmonella, Staphylococcus, and Streptococcus.
• To identify optimal computational strategies for accurate GI detection in bacterial genomes.

Main Methods:

• Combined multiple genomic island-associated features for analysis.
• Applied and compared various machine learning algorithms, including decision trees (J48), adaBoost, bagging, multiboost, and random forest.
• Evaluated classification accuracy using five standard performance metrics on single-genus and mixed-genus datasets.

Main Results:

• The decision tree approach generally outperformed other machine learning methods in classifying genomic islands.
• Ensemble algorithms, built upon J48 decision trees, demonstrated improved classification accuracy compared to base classifiers.
• Ensemble methods showed enhanced performance across datasets for Salmonella, Staphylococcus, Streptococcus, and a combined dataset.

Conclusions:

• Decision tree-based ensemble algorithms provide accurate classification of genomic islands (GIs) and non-GIs.
• These ensemble methods are recommended for future analyses of GI data.
• A software package for GI detection is available for public use.