Bayesian Kernel-Based Modeling of Single and Multiple Genetic Pathways for Cancer

• Activity Number: 647
• Type: Topic Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: International Indian Statistical Association
• Author(s): Sounak Chakraborty*+
• Companies: University of Missouri-Columbia
• Keywords: semi-parametric model ; genetic pathways ; variable selection ; Bayesian method ; cancer ; Hilbert space

Abstract

In this paper, we propose Bayesian kernel based regression model for modeling genetic pathways and their role in cancer studies. In the proposed models, the multidimensional regression functions are estimated using the reproducing kernel Hilbert space (RKHS) theory. The developed model is then used to quantify the pathway effects of a single genetic pathway on cancer. It is also extended to model multiple pathways jointly by using an additive model framework. Where, each pathway effect is estimated nonparametrically with the help of separate functions from RKHS. Based on Bayes factor and marginal likelihood we have proposed a scheme for testing single pathway effect and multiple pathway effects on the prostate cancer indicators. Along with testing pathway effects we design a stochastic search technique to do a simultaneous gene selection and also pathway selection, that will help us to pinpoint the relevant genes and pathways related with a disease. The performance of our proposed models are validated using 2 simulation studies and 2 real life data sets.