A novel approach to modeling multifactorial diseases using Ensemble Bayesian Rule classifiers

Area of Science:

• Biomedicine
• Computational Biology
• Genomics

Background:

• High-dimensional 'omic' datasets are crucial for understanding complex disease mechanisms.
• Multifactorial diseases arise from numerous small-effect factors, posing challenges for traditional modeling.
• Bayesian Rule Learning (BRL) effectively models high-dimensional data but struggles with data fragmentation in multifactorial diseases.

Purpose of the Study:

• To enhance Bayesian Rule Learning (BRL) for modeling multifactorial diseases.
• To implement and evaluate ensemble model combination strategies with BRL.
• To introduce a visualization tool for interpreting ensemble models.

Main Methods:

• Implemented three ensemble strategies: Uniform Combination (UC), Bayesian Model Averaging (BMA), and Bayesian Model Combination (BMC).
• Combined these strategies with BRL, creating Ensemble Bayesian Rule Learning (EBRL).
• Developed the Bayesian Rule Ensemble Visualizing tool (BREVity) for model interpretation.

Main Results:

• EBRL models using UC and BMC demonstrated superior predictive performance compared to BMA and classic machine learning methods on 25 public gene expression datasets.
• BMC showed greater reliability than UC when the ensemble incorporated sub-optimal models.
• BREVity effectively visualizes and extracts key rule patterns from EBRL models.

Conclusions:

• EBRL offers a powerful tool for modeling multifactorial diseases from high-dimensional data.
• The combination of EBRL and BREVity provides both predictive accuracy and interpretability for researchers.
• This approach addresses limitations of standard BRL in complex disease modeling.