Fig. 3: Integrated modeling of complementary assays quantifies the impact of methylation and co-factors on TF binding.
a, Combinations of TFs assayed (top) and unified model learned by ProBound (bottom). The model consists of the inferred energy logos for the monomeric and dimeric complexes (motifs) and the (b) inferred binding cooperativity (y axis) between Hth and Exd:Ubx for different relative positions (x axis) and orientations (red: parallel; blue: anti-parallel) of the subunits. Disk areas proportional to the affinity of the strongest predicted sequence highlight the most stable configurations. The shaded region indicates overlapping motifs. Schematics (inset) illustrate two configurations indicated on the plot. c, Combinations of TFs and methylated/unmethylated libraries assayed (schematic); methylation-aware binding models (motifs) using the alphabet in Extended Data Fig. 4a; and the impact of meCpG on binding free-energy (plots; −ΔΔGCpG→meCpG/RT on y axis) as a function of position within the binding site (x axis). Half-disk areas are proportional to the maximum affinity when either CpG (white) or meCpG (black) is substituted at the corresponding position in the highest-affinity sequence and highlight positions with high-affinity methylation readout. d, Impact of substituting a CpG (white) or meCpG (black) at a specific position in the highest-affinity binding site as quantified using ChIP-seq data. Each pair of bars corresponds to a substitution at a specific position and to red arrows in c. Antibody symbols indicate respective immunoprecipitated factor. P values were computed using an F-test (one-sided, *** indicates P < 10−3; Methods and Supplementary Table 2). e, Same as c for data simultaneously measuring methylation readout for meCpG, 5hmC and 6mA modifications.
