Bayesian and Meta- Analyses of Cell-Cycle Gene Expression Data

Gene expression experiments conducted under a variety of conditions can allow for concurrent tests of more than one hypothesis. It is common for such experiments to be conducted independently by different researchers, using possibly different microarray platforms. In the second and fourth chapter of this thesis, we propose a differential meta-analytic procedure to pool the data from various sources and test the relative significance of the hypotheses under consideration. The specific application made in this thesis is to 10 time-course cell-cycle experiments on fission yeast S. Pombe (Oliva et al., 2005; Peng et al., 2005; Rustici et al., 2004), and the hypotheses of interest concern the question of differential expression and periodic regulation of genes. Besides addressing the above differential meta-analysis issue, we explore how time-course gene expression data can be used to test for periodicity. In this context, the commonly used procedures for testing include the Permutation test by de Lichtenberg et al. (2005) and the G-test by Fisher (1929), both of which are designed to evaluate periodicity against noise; however, it is possible that a given gene may have expression that is neither cyclic, nor just noise. In the third chapter, we introduce an Empirical Bayes approach to test for periodicity and compare its performance in terms of sensitivity and specificity with that of the other two methods through simulations and by application to the S. Pombe cell-cycle gene expression data. We use ‘conserved’ and ‘cycling’ genes by Lu et al. (2007) to assess the sensitivity, and CESR genes by Chen et al. (2003) to assess the specificity of our method. Kocak, M., Zhang, G., Narasimhan, G., George, E.O., Pyne, S. (2010) use George and Mudholkar’ (1983) ‘Difference of Two Logit-Sums’ method to pool bivariate P-values across independent experiments, assuming independence within a pair. We propose a Bayesian approach for pooling bivariate P-values across independent experiments, which accounts for potential correlation between paired P-values. We will investigate the operating characteristics of the Bayesian method trough simulations and apply it to the S. Pombe cell-cycle data