4. Conclusions
In prospect of the presented results it becomes clear that biological interpretations based on expression changes in large data sets might be misleading because they potentially interfere with technical effects. In particular variation of RNA quality and RNA quantity constitute major sources of technical bias and should be considered in every large-scale gene expression study as well as in integrative analyses of combined data sets.
What can be done to alleviate unwanted technical variation and its effects? The importance of quality control for microarray gene expression experiments is already well accepted—partly thanks to initiatives such as the MAQC. Independent measurements, for example of RNA quality, are de-facto standard protocol. These measures ensure that all samples fulfill at least a minimum quality criterion prior to microarray hybridization. We here however showed how several technical factors affect expression results to a degree that goes beyond simple good-quality vs. bad-quality decisions. We therefore encourage that technical variation be explicitly considered at the various steps of a microarray experiment (during sample preparation and microarray hybridization in the lab as well as during data analysis procedures) as part of a comprehensive quality control.