(This might need to be moved to the methods or the beginning of the next section)
TODO: get clusters from consensus clustering and see if there is enrichment in GO terms
Figure S3: Consensus clustering (DONE)
Figure S4: Expression of genes that come from consensus clustering. (DONE)

An early morning gene regulatory network

Given this set of genes, we wished to identify possible regulatory links by inferring a gene regulatory network.  Because genes change their expression rapidly at dawn, it is likely that there is a time delay between the time the gene is expressed and the time the protein is expressed.  For this reason, we applied a time delay network inference method called dynGenie3.  This analysis predicted a dense hairball network (Figure S5– not done ) that had qualities of a scale-free network (Figure S6 ).  Although there was some spatial localisation by time of peak expression (Figure S7 ) and GO term (Figure S8 ), it is hard to interpret such a dense network.
To be able to interpret this dense network, we wished to determine whether genes associated with certain GO terms were found upstream or downstream of genes with other GO terms (Figure 3A ).  We compared the observed frequency of regulatory relationships between genes with certain GO terms in our predicted network with the expected frequencies, calculated by repeatedly shuffling the GO terms (Figure 3A ).  This would help us learn how pathways related to light, circadian, and temperature interact with one another.  Within our inferred network, only genes that are labelled ‘DNA binding’ can regulate other genes, so this GO term was not included in the analysis unless it was the only GO term associated with a given gene.  In the network that only includes DNA binding genes, we observe that circadian, light, phosphorylation and signalling genes tend to be upstream regulators, while temperature and growth tend to be downstream (Figure 3B ).  Interestingly, we see that abscisic acid (ABA) genes tend to be upstream of cold, water, photosynthesis and biosynthesis genes– this matches well with the known role of ABA signalling (Daszkowska-Golec 2016).
It is also possible to depict these results as a network that only includes edges where there is a p-value<0.05.   We developed these networks for all genes (Figure 3C ) and for DNA binding genes (Figure 3D ).  hormone-related GO terms are more well-connected in the network that includes all genes than the network that only includes DNA binding proteins, suggesting that hormone-related DNA binding genes may regulate genes that do not have DNA binding capacity (Figure 3C ).  Within the DNA binding-only network, the circadian genes seem to interact with light and temperature genes, but there is no enrichment between light and temperature directly (Figure 3D ) .