Supplementary Figure 15: Weak interactions have a predominant effect on expression level.
From: Deciphering eukaryotic gene-regulatory logic with 100 million random promoters
(a,b) Distinct regulatory networks for two sequences with nearly identical expression levels. Model-predicted regulatory networks for two 80 bp native promoters fragments using the positional pTpA+glucose model. Purple nodes: Individual TFs. Edge thickness and spacing: interaction strengths for each TF (green: positive/activating; red: negative/repressing). Both 80 bp promoter sequences have intermediate measured expression levels of ~10.5 in pTpA+glucose, but different regulatory networks: (a) the -160:-80 region of YDR456W is dominated by a single strong interaction, while (b) the −320:-240 region of YIR035C is regulated by many weak interactions. (c) Abundant weak regulatory interactions also explain most of expression in the most relevant promoter region. Cumulative distributions (y axis) of the number of regulatory interactions (black) and fraction of expression explained (that is cumulative total/sum of all effect sizes; red) for each interaction strength (x axis). Here, we consider only native promoter fragments from the -160:-80 region (relative to the TSS), which is the region our models are trained on and corresponds to the most active promoter region (Supplementary Fig. 5). The magnitude (and not the sign) of the interaction strength is considered.
