Data normalization: a global normalization was performed across all data from a given microarray platform (see Methods section).
Each of the 16 pools was then independently hybridized to an Amby_001 custom Affymetrix GeneChip (for a more detailed description of the microarray platform see [ 31] and [ 32]).
RNAseq is a more powerful technique than microarrays, and relies on a different experimental/statistical approach (for a comparison between microarray and RNAseq platforms see Guo et al. [ 53]), which may hinder comparisons among studies.
Quantitative reverse transcription PCR (qRT-PCR) was used to validate the results of the different microarray platforms [see Additional file 2].
Because microarray platforms use their own element IDs rather than common gene IDs, two search tools were developed to determine the relationship between microarray elements and rice genes.
To convert from a source to a target microarray platform, we chose to model each individual sequence tag in the target platform as a linear combination of all sequence tags from the source platform (see Fig. 1 and Supplementary Methods).
Consistent with the relatively small degree of overlap between the microarray platforms (< = 8%, see the Data Processing section in the Text S1 for details), the representative models developed from the two independent datasets have no genes in common.
As expected, there are no commonly deregulated genes resulting from the intersection of all five atrophy profiles; this is probably due to the experimental differences between studies included in the work (stimuli inducing atrophy, microarray platforms, muscle types, and organisms, see Table 1).
For example, the correlation coefficient between two series of locus-matched values obtained by distinct Authors, using different microarray platforms (samples A1 and A3, respectively; see Table 1), was 0.23 in absence of any inter-sample normalization, 0.34 following standard quantile normalization for all values and 0.41 following scaled quantile method.
