Abstract

We present a Bayesian process-based model for spotted microarray data incorporating available information about the experiment from target gene preparation to image analysis. We demonstrate that, using limited calibration data, our method can estimate absolute gene concentrations from spotted microarray intensity data. Number of transcripts (copies of the gene sequence) per microgram total RNA are obtained for each gene, enabling comparisons of transcript levels within and between samples. All parameters are estimated in one Markov chain Monte Carlo run thereby propagating uncertainties throughout the model. We reparameterize the core of the model, binomial selection, and show identifiability of the parameters. Using a small data set, we illustrate potentials of our method discriminating it from conventional, ratio-based methods. This chapter gives a thorough description of the statistical methodologies that form the foundation of the biology-focused companion paper.

Introduction

Analysis of microarray data is challenging due to the huge number of measurements made in each experiment and the large uncertainty associated with it. When spotted cDNA microarrays are used, gene expression levels are generally measured as a log-ratio of the fluorescence intensity of two cDNA samples to reduce systematic effects in the data, though biological information that lies in the absolute concentrations may be lost. Samples are derived from mRNA by reverse transcription and dye labeling and cohybridized to an array of DNA probes on a microscope slide. The intensities are measured by imaging the array in an optical scanner. There are several sources of variation associated with each step in the experimental procedure that influence the measured intensities, and hence the expression.