Variation in fiberoptic bead-based oligonucleotide microarrays: dispersion characteristics among hybridization and biological replicate samples

BACKGROUND: Gene expression microarray technology continues to evolve and its use has expanded into all areas of biology. However, the high dimensionality of the data makes analysis a difficult challenge. Evaluating measurements and estimating the significance of the observed differences among samples remain important issues that must be addressed for each technology platform. In this work we use a consecutive sampling method to characterize the dispersion patterns of data generated from Illumina fiberoptic bead-based oligonucleotide arrays. RESULTS: To describe general properties of the dispersion we used a linear function SD = a + bY(mean), approximating the standard deviation across arrays (Y(mean )is the mean expression of a given consecutive sample). First we examined three levels of variability: 1) same cell culture, same reverse transcription, duplicate hybridizations; 2) same cell culture, reverse transcription replicates; 3) parallel cultures. Each higher level is expected to introduce a new source of variability. We observed minor differences in the constant term: the mean values are 3.5, 3.1 and 3.5, respectively. However, the mean coefficient b increased from 0.045 to 0.147 and 0.133. We compared the coefficients derived from the consecutive sampling to those obtained from the standard deviation of individual gene expressions and found them in good agreement. In the second experiment samples we detected 11 genes with systematically different expressions between the experiment samples treated with glucose oxidase and controls and corroborated the selection using the Mann-Whitney and other tests. We also compared the consecutive sampling and coincidence method to t-test: the average percentage of consistency was above 80 for the former and below 50 for the latter. CONCLUSION: Our results indicate that the consecutive sampling method and standard deviation function provide a convenient description of the overall dispersion of Illumina arrays. We observed that the constant term of the standard deviation function is at average approximately the same for duplicate hybridization as for the assays with additional sources of variability. Furthermore, among the genes affected by glucose oxidase treatment we identified 6 genes in oxidative stress pathways and 5 genes involved in DNA repair. Finally, we noted that the consecutive sampling and coincidence test provide, under given conditions, more consistent results than the t-test. REVIEWERS: This article was reviewed by Alexander Karpikov (nominated by MarkGerstein), Jordan King and Eugene V. Koonin.