Oligonucleotide microarray data distribution and normalization

Abstract

Variations in oligonucleotide microarray probe signals that result from various factors, including differences in sample concentrations, can lead to major problems in the interpretation of data obtained from different experiments. Normalization of such signals is typically performed by procedures involving division by a constant approximately determined by average signal intensities as, e.g., in the Affymetrix software. Here we show that Affymetrix oligonucleotide probe signal distributions can be fitted by using a superposition of two normal or two extreme distributions, and that by using such distributions we can normalize data with high accuracy (parametric algorithm). We also developed a second algorithm (nonparametric) based on ranking of signal intensities which gave equal or better normalization than the parametric one. These approaches have been used for normalization of three sets of data obtained from cancer cell lines, peripheral blood mononuclear cells from patients with HIV infections, and adipose cells from patients with diabetes, and others. Both, parametric and nonparametric normalization procedures, were found to be superior when compared to the standard global normalization approach [Affymetrix Microarray Suite User Guide. Version 4.0 (2000)]. These results suggest that the new approaches may be helpful for microarray data normalization especially for comparison of clinical data where interpatient differences can be large and difficult to avoid.