SurvBART

SurvBART implements a fully Bayesian sum-of-trees ensemble to predict patient survival outcomes from high-dimensional gene expression data.

Key Features:

• Bayesian hierarchical approach: Employs a fully Bayesian hierarchical framework to model uncertainty and incorporate prior information in survival prediction.
• Sum-of-trees model: Uses a non-parametric 'sum-of-trees' ensemble that aggregates tree-based components and explicitly integrates three popular survival models.
• Non-parametric methodology: Avoids parametric distributional assumptions to provide flexibility in modeling complex gene expression effects.
• Additive and interaction effects: Captures both additive effects and gene–gene interactions for improved predictive accuracy.
• Model-free variable selection: Performs model-free variable selection with control of false discovery rates to identify important prognostic genes.
• Performance on microarray data: Demonstrates competitive predictive performance on simulated and real microarray datasets.

Scientific Applications:

• Oncology prognostication: Predicting patient survival and enabling risk stratification from high-dimensional tumor gene expression profiles.
• Biomarker discovery: Identifying prognostic gene markers for downstream biological validation and clinical decision-making.
• Personalized treatment strategies: Informing personalized treatment approaches by linking gene expression patterns to patient outcomes.
• Method benchmarking: Benchmarking and comparing predictive performance across simulated and real microarray datasets.

Methodology:

Implements a fully Bayesian hierarchical, non-parametric 'sum-of-trees' ensemble that integrates three survival models, captures additive and interaction effects, and performs model-free variable selection with false discovery rate control, evaluated on simulated and real microarray datasets.