Improved Microarray Spot Segmentation by Combining two Information Channels

High-throughput gene expression is an important aspect of modern post-genomic research. Microarray technology is the driving force of this revolution, a technology that allows the simultaneous monitoring of expression for thousands of genes. The need for accurate and reproducible research has driven the development of robust analysis frameworks for maximizing the information content of biological data. In microarray imaging technologies, several non-linearities in the experimental process render the measured expression values prone to variability and often, to poor reproducibility. Accurate segmentation of the true signal is a very important task, not least because a single value per spot needs to be derived for further knowledge discovery analysis. In this paper, we present a fully automatic segmentation method for improving the spot segmentation result. The method doesn't make any assumptions concerning the number of classes present in each image spot, and it isn't driven only by the most intense features, since it takes into account the underlying "hybridization ground truth" derived from both information channels of the spotted arrays. Our method is compared to widely used, state-of-the-art segmentation methods in microarray image analysis in a study of a metabolic disorder in yeast, where replicates of reporters are present. Initial results indicate that our method yields more reproducible log ratio measurements across replicates