Iterative Bayesian Model Averaging: a method for the application of survival analysis to high-dimensional microarray...

Area of Science:

• Bioinformatics
• Genomics
• Cancer Research

Background:

• Microarray technology is crucial for identifying cancer biomarkers.
• Iterative Bayesian Model Averaging (BMA) algorithm was previously developed for classification.
• This study extends the iterative BMA algorithm for survival analysis on high-dimensional microarray data.

Purpose of the Study:

• To develop a highly predictive model for patient time-to-event outcomes using selected genes from microarray data.
• To combine multiple models effectively by averaging their posterior probability distributions.
• To achieve high prediction accuracy with a minimal set of genes for cost-effective cancer prognostics.

Main Methods:

• Application of the iterative BMA algorithm to breast cancer and diffuse large B-cell lymphoma (DLBCL) datasets.
• Selection of predictor genes and models from training data.
• Division of patients into high- and low-risk categories in validation datasets using selected genes and models.

Main Results:

• For breast cancer data, 15 genes across 84 models were selected, yielding distinct risk groups (p=7.26e-05). Comparable results were achieved with the top 5 genes.
• For DLBCL data, 25 genes across 3 models were selected, also resulting in significantly distinct risk groups (p=0.00139).
• The algorithm demonstrated high prediction accuracy and selected a small, cost-effective number of genes.

Conclusions:

• The iterative BMA algorithm for survival analysis effectively handles model uncertainty.
• The procedure outperforms other methods in predictive performance.
• It offers a highly accurate and cost-effective prognostic tool for clinical settings.