Supplementary Figure 4: Predicting expression from TF binding.
From: Deciphering eukaryotic gene-regulatory logic with 100 million random promoters
(a) Relationship between predicted binding of individual TFs and expression level. Measured expression level (pTpA+Glu data; y axis) vs. predicted binding (x axis) for Abf1 (left), Rsc30 (middle), and Ume6 (right). Ume6 (a similar motif to Rsc30), is positively correlated with expression overall (r=0.14), but has a strong negative trend at high predicted occupancies. (n = 250,000 promoters for each; Pearson’s r shown at bottom right; Top: Motifs. Blue lines: GAM lines of best fit. Gray shaded areas: 95% confidence intervals.) (b) Model correctly predicts expression of designed sequences. Shown are the model-predicted (x axis) and measured (y axis) expression levels for 1,000 in silico generated random sequences designed based on the pTpA+glucose billboard model’s predictions to span a range of expression levels, and then measured experimentally. (n = 1,000 promoters; Pearson’s r2 shown at bottom right.).
