Technical variation in whole-genome, single-probe, one-color microarrays has some common properties that are the subject of this paper. One such property, which is evidenced in measurement of different materials in the same laboratory, involves non-linearity in probe responses. For all microarray types, non-linearity in probe responses exists but is similar in magnitude to the random variation that would occur were there no non-linearity. For this reason, technical replicates seem to have value in prediction of the magnitude of within-laboratory variation and of the uncertainty in differential expression determinations. In this paper, this non-linearity is examined on the basis of the regression lack-of-fit statistic appropriate for testing whether the reference materials in the MAQC study differ according to the mixture proportions set in the experimental design. Another property, which is evidenced in measurement of the same material in different laboratories, involves a source of variation that affects many probes. For all the microarray types, such sources are appreciable. These sources induce statistical dependence among measurements for different genes. In this paper, such sources are characterized through application of factor analysis accompanied by factor rotation. This characterization shows the limitations in the use of array normalization as an approach to reducing inter-laboratory differences. Overall, this paper shows that the platforms considered are comparable in terms of the properties investigated.
Citation: Nature Biotechnology
Volume: 111 No. 5
Pub Type: Journals
Biomarker, gauge R&R, inter-laboratory study, microarrays, response linearity, sources of variation