Bioinformatics toolbox

We trained a random forest (RF) classifier to predict whether a TF binding site is a direct or indirect site using the proximal binding of other TFs in the co-binding region. We used the TreeBagger implementation of RF in the MATLAB software (MATLAB and Bioinformatics Toolbox Release 2015b, The MathWorks, Inc., Natick, Massachusetts, United States). More specifically, using the region-TF matrix (159,204 × 167), we took the rows that contained either direct or indirect binding sites of the TF, used the columns corresponding to the direct or indirect binding of the TF as the prediction target and the rest of the columns (binding of the other TFs) as the features. For each sequence specific co-binding TF, the dTF and iTF columns were combined, ignoring the motif information. Therefore, the prediction is harder because it based only on the identity but not the motif information of the co-binding TFs. For each sequence-specific TF, we trained five RFs, each with a distinct random subset (80%) of the data, and then tested on the rest of 20% data. The prediction accuracy values of the five classifiers are then averaged for each TF. The correlation between direct and indirect binding of a TF (shown in Fig. 5c) were computed using the corresponding TF vectors in the module-TF matrix.

The diurnal expression data with 4-h intervals for Arabidopsis thaliana were obtained from Mockler et al.^{75 (link)} and adjusted to 2-h interval time series by interpolation using the SRS1 cubic spline function (http://www.srs1software.com/). The diurnal expression data with 2-h intervals for K. fedtschenkoi was generated in this study. The diurnal expression data with 2-h intervals for Ananas comosus was obtained from Ming et al.^{4 (link)}. The gene expression data were normalized by Z-score transformation. The hierarchical clustering of gene expression was performed for genes in each ortholog group using the Bioinformatics Toolbox in Matlab (Mathworks, Inc.) based on Spearman correlation (Supplementary Method 14).

To identify phosphopeptides with significant changes in abundance relative to the 0 s LightR-Src condition, we utilized paired Student’s t-test (p-value<0.05). Additionally, to ensure phosphopeptide changes were not simply background, we filtered for peptides with average abundances (n = 3) that were >1.4 relative to the 0 s LightR-Src condition. The average abundance for each phosphopeptide was log2-transformed and visualized using MATLAB (version R2019b, Bioinformatics Toolbox version 4.13, MathWorks). Data were plotted with the ‘clustergram’ function with hierarchical clustering using Euclidean distance.