Figure 2

Comparison of three circadian microarray datasets. The power to detect circadian genes is greatly increased when independent datasets are combined. (a) The degree of circadian regulation of the Arabidopsis genome as originally reported in different studies [4, 6, 7]. (b) The number of unique unreplicated time series (generated by random shuffling of Harmer technical replicates) that identifies each of the circadian-regulated genes found in at least one shuffled time series. The shaded portion indicates the genes that are found to be circadian in a majority of the time series. (c) The shuffled Harmer datasets were analyzed according to the parameters originally used for the Covington dataset; only genes common to the two microarray platforms were considered. (d) The Covington dataset was reanalyzed according to the parameters originally used for the Edwards dataset, with the exception that only genes expressed in both datasets were evaluated. Also shown are the results of the analysis of the combined Covington and Edwards datasets, as well as the Michael datasets. For the individual and combined Covington plus Edwards datasets, only genes that are expressed in both of the individual data sets are considered. (e) The unions and intersections of sets of genes determined to be circadian expressed by the different datasets. Harmer-A and Harmer-B represent the two of the 20 shuffled datasets with the degree of circadian regulation closest to the 50th percentile. The percent overlap for each pair is shown in parentheses. (f) There is substantial overlap in the identity of circadian regulated genes (shown as numbers within Venn diagram circles) found by the three combined Covington plus Edwards datasets. The number in the lower right represents the number of genes that are expressed in both the Covington and Edwards datasets. (g) Collections of circadian genes identified in different datasets share substantial identity with the circadian genes found by each of the three combined Covington and Edwards datasets.