BAYESIAN VARIABLE SELECTION FOR SURVIVAL DATA USING INVERSE MOMENT PRIORS.

Area of Science:

• Genomics
• Biostatistics
• Cancer Research

Background:

• High-dimensional cancer genomic studies require efficient variable selection for gene discovery and treatment response prediction.
• Censored survival data is commonly encountered in cancer research, posing analytical challenges.

Purpose of the Study:

• To introduce a novel Bayesian variable selection procedure for high-dimensional genomic data with censored survival outcomes.
• To develop and implement an R package (BVSNLP) for this procedure, enabling parallel computing and efficient model space exploration.

Main Methods:

• A Bayesian variable selection approach utilizing a mixture prior (point mass at zero and inverse moment prior).
• Integration with the Cox proportional hazard model's partial likelihood.
• Implementation in the R package BVSNLP with stochastic search and Bayesian model averaging for prediction.

Main Results:

• The proposed Bayesian algorithm demonstrated superior performance in simulation studies compared to existing variable selection methods.
• Consistent variable selection was observed when applied to real-world cancer genomic datasets.

Conclusions:

• The developed Bayesian procedure offers an effective tool for variable selection in cancer genomics.
• The BVSNLP package facilitates robust gene discovery and prediction of treatment response in complex cancer datasets.