BoNB

BoNB applies an ensemble of Naïve Bayes classifiers with bootstrap aggregating to analyze genome-wide single nucleotide polymorphism (SNP) data for genetic marker selection and subject classification in multifactorial disease studies.

Key Features:

• Bootstrap aggregating of an ensemble of Naïve Bayes classifiers: Uses bagging to create an ensemble of Naïve Bayes classifiers that reduces variance and mitigates overfitting in high-dimensional genomic SNP data.
• Attribute ranking and selection strategy: Implements a strategy for ranking and selecting SNP attributes so that each classifier in the ensemble uses the most informative genetic markers.
• Permutation-based procedure for biomarker significance: Employs a permutation-based procedure to assess the statistical significance of biomarkers based on their marginal utility for classification.

Scientific Applications:

• Genetic marker selection from GWAS: Selects significant SNP biomarkers from genome-wide association studies (GWAS) for multifactorial diseases.
• Subject classification in multifactorial diseases: Classifies subjects using genome-wide SNP profiles to support studies of diseases such as Type 1 Diabetes.
• Analysis of genetic and environmental interactions: Analyzes complex interaction patterns between genetic traits and environmental factors by simultaneously considering all SNPs to better capture heritability components.

Methodology:

Builds an ensemble via bootstrap aggregating of Naïve Bayes classifiers, applies an attribute ranking and selection strategy per classifier, uses a permutation-based procedure to evaluate biomarker significance, and was evaluated on datasets such as the Wellcome Trust Case-Control study on Type 1 Diabetes with comparative analyses versus standard Naïve Bayes and HyperLASSO penalized logistic regression.