iterativeBMAsurv

iterativeBMAsurv applies Bayesian Model Averaging to perform survival analysis on high-dimensional microarray data and identify minimal predictive gene sets for cancer prognosis and risk stratification.

Key Features:

• Bayesian Framework: Utilizes Bayesian Model Averaging to manage model uncertainty by computing weighted averages of posterior probability distributions across competing models.
• Variable Selection: Selects a minimal, highly predictive set of genes associated with patient-specific time-to-event outcomes such as death, relapse, or metastasis.
• High Predictive Accuracy: Demonstrates high predictive accuracy on breast cancer and diffuse large B-cell lymphoma (DLBCL) microarray datasets while consistently selecting small subsets of predictor genes.
• Risk Stratification: Applies selected genes and models from training data to validation datasets to stratify patients into risk groups, with separation assessed by log-rank tests and associated p-values.
• Cost-Effectiveness: Can achieve comparable predictive results using only the top selected genes with 100% posterior probabilities, reducing the required biomarker set.

Scientific Applications:

• Prognostic Biomarker Discovery: Identifies key prognostic genes from microarray data that associate with survival outcomes.
• Personalized Medicine: Enables patient-level risk stratification to inform tailored therapeutic strategies.
• Clinical Trial Design: Supports identification and enrichment of relevant patient subgroups for clinical trials using selected predictive gene sets.

Methodology:

Performs a multivariate procedure that identifies competing models from high-dimensional microarray data, computes weighted averages of their posterior probabilities via Bayesian Model Averaging, and applies the selected genes and models to validation datasets for risk stratification assessed by log-rank tests.