Statistical Methods in Cancer Genomics.

Abstract

Genomic and proteomic experiments have become widely applied in cancer profiling studies over the past decade. The genomics era is marked by the success of using DNA microarrays to delineate genome-scale gene expression patterns to pinpoint disease mechanism at the molecular level. An increasing number of studies have profiled tumor specimens using distinct microarray platforms and analysis techniques. With the accumulating amount of microarray data, integrative analysis has the potential to identify common gene expression patterns across data sets and tissue types. In this proposal, I introduce a Bayesian mixture model-based approach for meta-analysis of microarray studies. A probabilistic measure of gene differential expression is used as a scaleless quantity for an integrative analysis of DNA microarray data sets across platforms and laboratories. The role of DNA microarrays has been primarily on the discovery side to screen through thousands of genes for potential disease biomarkers. In this respect, Tissue Microarrays (TMAs) have provided a proteomic platform for downstream validation studies of these target discoveries. The other part of this proposal concerns an implementation of measurement error models for patient survival outcome analysis using TMA expression data. Two goals are explored: 1) in a two-stage approach, a Latent Expression Index (LEI) is introduced as a summary index for the TMA repeated expression measures; 2) a joint model of survival and TMA expression data is established via a shared random effect. Bayesian estimation is carried out using a Markov Chain Monte Carlo (MCMC) method. As an extension to the measurement error models, I further propose a Cell Mixture model to allow a wider range of inferences for TMA expression data.