**Workflow for the identification of candidate cyclic genes.** (**A**) A primary analysis was first carried out to identify candidate cyclic genes based on the periodicity of their expression profile in the microarray series. A permutation algorithm that compares the experimental datasets with randomized datasets was used to identify the cyclic genes in the three species. Local time windows (groups shown in Fig. S2 in the supplementary material), within which the exact ordering of the samples based on the phase of the oscillation is hard to ascertain, were first defined. Then, 10^{4} permutations within these defined windows were created for each dataset. We next applied the Lomb-Scargle (L-S) algorithm, which is a Fourier-related analysis, to each permutation to rank the candidate cyclic genes based on their periodicity. A global ranking from the most to the least cyclic probeset was computed using the product of the ranks of the 10^{4} orderings. A primary filtration was used to remove the non-expressed and non-reliably detected probesets. Finally, the distributions of *P*-values from the experimental and random datasets were compared using a *t*-test. The number of candidate cyclic genes was determined by the first rank where the *t*-test *P*-value was higher than 0.01. (**B**) The secondary analysis for the mouse dataset selects the candidate cyclic genes that are in common between the two mouse microarrays. For chicken and zebrafish, we used as filtration criteria the signal thresholds of known cyclic genes. (**C**,**D**) Venn diagrams illustrate the number and overlap of candidate cyclic genes after primary (C) and secondary (D) analysis.

## PubMed Commons