The microarray research community and industry got a big boost last week with the release by the
Microarray Quality Control (MAQC) Project Consortium, of a massive collection of data attesting to the reproducibility and reliability of microarray-based gene-expression profiling.
Attempting to lay to rest, once and for all, the question of
microarray data reliability and reproducibility the Consortium, representing 137 scientists from 51 academic, government, and industry institutions, compared data from two standard RNA samples on seven array platforms and three non-array platforms. Team members cross-checked their numbers between experiments, across sites, and across platforms, accounting for nearly every variable imaginable, from external RNA controls, to one-color versus two-colors, to statistical methods. The results were published in the September issue of
Nature Biotechnology, which devoted
67 pages to the material, including an editorial, a foreword, three commentaries, four analyses, and two research articles.
The main conclusion: consistency across sites and between platforms is generally high. For instance, "all one-color microarray platforms had a median CV [coefficient of variance] of 5-15% for the quantitative signal and a concordance rate of 80-95% for the qualitative detection call between sample replicates." That?s pretty good, but it means that even under the best circumstances, gene lists will still differ somewhat from person to person and place to place. In the hands of the MAQC, "Lists of differentially expressed genes averaged
89% overlap between test sites using the same platform and 74% overlap across one-color microarray platforms." Individual researchers, trudging along on a particular problem, might not fare quite so well.
It probably makes sense at this point to focus on pathways and broad functional relationships, rather than on individual genes. Nevertheless, basic researchers can largely stop worrying that their colleagues will fail to reproduce their findings ? even if they use different platforms (say, Affymetrix versus Agilent or GE Healthcare) or assays (TaqMan assays as opposed to microarrays). More significantly, drug developers can begin using gene expression microarray data to bolster drug applications, and regulatory agencies can understand the value of that data.
We'll be looking into the MAQC project in depth in our November issue.