MOTIVATION: The radioactivity labeled DNA array platform is a robust and accurate way for a high-throughput measurement of gene expression levels in biological samples. Despite its high degree of sensitivity and reproducibility, this platform has several sources of variation. These are related to the presence of saturation effects in the array images and impede the degree of accuracy at which gene expression levels are determined. RESULTS: Here we describe a simple, but effective, approach for combining expression data from a series of autoradiographic exposures of variable length. This technique increases the sensitivity of this array platform by detecting low-expressed genes at longer exposures. It also improves the measurement accuracy of highly abundant genes by considering only values from the linear portion of dependency between the exposure times and gene intensities. As a result, the described approach improves the outcome of the subsequent steps of array data normalization and mining.

MOTIVATION: Detailed comparison and analysis of the output of DNA gene expression arrays from multiple samples require global normalization of the measured individual gene intensities from the different hybridizations. This is needed for accounting for variations in array preparation and sample hybridization conditions. RESULTS: Here, we present a simple, robust and accurate procedure for the global normalization of datasets generated with single-channel DNA arrays based on principal component analysis. The procedure makes minimal assumptions about the data and performs well in cases where other standard procedures produced biased estimates. It is also insensitive to data transformation, filtering (thresholding) and pre-screening.