How high is the level of technical noise in microarray data?

Lev Klebanov¹^,2 and Andrei Yakovlev¹^,2

¹Department of Probability and Statistics, Charles University, Sokolovska 83, Praha-8, CZ-18675, Czech Republic

²Department of Biostatistics and Computational Biology, University of Rochester, 601, Elmwood Avenue, Box 630, Rochester, New York 14642, USA

author email corresponding author email

Biology Direct 2007, 2:9doi:10.1186/1745-6150-2-9


Published:	11 April 2007

Abstract

Background

Microarray gene expression data are commonly perceived as being extremely noisy because of many imperfections inherent in the current technology. A recent study conducted by the MicroArray Quality Control (MAQC) Consortium and published in Nature Biotechnology provides a unique opportunity to probe into the true level of technical noise in such data.

Results

In the present report, the MAQC study is reanalyzed in order to quantitatively assess measurement errors inherent in high-density oligonucleotide array technology (Affymetrix platform). The level of noise is directly estimated from technical replicates of gene expression measurements in the absence of biological variability. For each probe set, the magnitude of random fluctuations across technical replicates is characterized by the standard deviation of the corresponding log-expression signal. The resultant standard deviations appear to be uniformly small and symmetrically distributed across probe sets. The observed noise level does not cause any tangible bias in estimated pair-wise correlation coefficients, the latter being particularly prone to its presence in microarray data.

Conclusion

The reported analysis strongly suggests that, contrary to popular belief, the random fluctuations of gene expression signals caused by technical noise are quite low and the effect of such fluctuations on the results of statistical inference from Affymetrix GeneChip microarray data is negligibly small.

Reviewers

The paper was reviewed by A. Mushegian, K. Jordan, and E. Koonin.