An Arabidopsis gene expression predictor enables inference of transcriptional regulators

Gene expression regulation by transcription factors (TF) has long been studied, but no model exists yet that can accurately predict transcriptome profiles based on TF activities. We have constructed a universal predictor for Arabidopsis to predict the expression of 28192 non-TF genes using 1678 TFs. Applied to bulk RNA-Seq samples from diverse tissues, the predictor produced accurate predicted transcriptomes correlating well with actual expression, with average correlation coefficient of 0.986. Having recapitulated the quantitative relationships between TFs and target genes, the predictor further enabled downstream inference of TF regulators for genes and pathways, i.e. those involved in suberin, flavonoid, glucosinolate metabolism, lateral root, xylem, secondary cell wall development, and endoplasmic reticulum stress response. Our predictor provides an innovative approach to study transcriptional regulation.