A0629
Title: Semiparametric Bayesian kernel survival model using multilevel learning
Authors: Inyoung Kim - Virginia Tech (United States) [presenting]
Abstract: Motivated by a breast cancer gene-pathway data set, which exhibits the "small n, large p" characteristics, a semiparametric variable selection method is proposed for the Bayesian kernel survival model to simultaneously study the effects of both clinical covariates and gene expression levels within a pathway on survival time and also identify important variables associated to survival time. The unknown high-dimension functions of pathways are modeled via the Gaussian kernel machine to consider the possibility that genes within the same pathway interact with each other. To address the multiple comparisons problem under a full Bayesian setting, a similarity-dependent procedure is proposed based on the Bayes factor to control the family-wise error rate. The outperformance of the approach is demonstrated under various simulation settings and pathways data.
