Sir

Michael Eisenstein, in his Technology Feature “Quality control” (Nature 442, 1067–1070; 2006), reports recent improvements in microarray technologies. He discusses the establishment of systems for “clearly defining how [microarray] data were obtained”, such as the MIAME (minimum information about a microarray experiment) annotation standards.

There are two problems inherent to the use of such standards. First, microarrays generate datasets whose intrinsic dimensionalities are several orders of magnitude greater than the information content of the standards. Second, one cannot know a priori all the biological factors that might affect the results reported by microarrays.

In terms of the first point, the prospect of defining 30,000-odd different MIAME descriptors — the routine size of micro-arrays — would be daunting under any circumstances. The second point is neatly illustrated by work described in the same issue (J. Molinier, G. Ries, C. Zipfel and B. Hohn Nature 442, 1046–1049; 2006). This study reported the unexpected finding of transgenerational, non-genetic memory of stress events in the flowering plant Arabidopsis thaliana. Although transcriptome analyses in this case did not detect differences in global gene expression (possibly due to lack of sensitivity), the question remains as to whether one should be required to record the prior generational treatments of plant lines in microarray experiments — or in any experiment. Such a requirement appears equally daunting.

How should we proceed? Reducing the costs of microarray technology so that experiments can be readily reproduced across laboratories seems a reasonable approach. Relying on minimal standards of annotation such as MIAME seems unreasonable, and should be abandoned.