Robust preprocessing of gene expression microarrays for independent component analysis

Abstract

Oligonucleotide Microarrays are useful tools in Genetic Research as they provide parallel scanning mechanisms to detect the presence of genes using test probes composed of controlled segments of gene code built by masking techniques. The detection of each gene depends on the multichannel differential expression of perfectly matched segments (PM) against mismatched ones (MM). This methodology, devised to robustify the detection process poses some interesting problems under the point of view of Genomic Signal Processing, as test probe expressions are not in good agreement with the proportionality assumption in most of the genes explored. These cases may be influenced by unexpected hybridization dynamics, and are worth of being studied with a double objective: gain insight into hybridization dynamics in microarrays, and to improve microarray production and processing as well. Recently Independent Component Analysis has been proposed to process microarrays. This promising technique requires the pre-processing of the microarray contents. The present work proposes the de-correlation of test probes based on probe structure rather than the classical "blind" whitening techniques currently used in ICA. Results confirm that this methodology may provide the correct alignment of the PMMM pairs maintaining the underlying information present in the probe sets. © Springer-Verlag Berlin Heidelberg 2006.