Fig. 3: IQ use of low-affinity binding, spatial preferences, and transcription factor interactions.

A Density plots showing the distribution of observed AP for loci grouped into seven bins based on their IQ-predicted dAP on the epiblast to mesoderm mouse trajectory. Gray dots represent all loci in the dataset, while colored dots represent the density of loci in the specific bin of each plot. The gradual shift in distributions demonstrates IQ’s ability to distinguish between strongly and weakly induced/repressed cis-regulatory elements (CREs). B Comparison of model performance (R²) when using IQ’s integrated motif energy approach (methods) versus considering only the top binding site for each CRE, across different mouse gastrulation trajectories. C Comparison of the impact of low-affinity versus high-affinity binding sites. Y-axis: mean dAP for loci with multiple low-affinity sites minus the mean dAP for loci with a single high-affinity site (methods). X-axis: residual R² of each motif in the trajectory model. Each point represents one mouse gastrulation trajectory. D Heatmaps showing the inferred spatial binding preferences for transcription factor models in a mouse (left) and human (right) trajectories. Columns represent different distances from the CRE center. E Comparison of model performance (R²) with and without epigenomic features for different trajectories in mouse gastrulation. F Residual R² values showing the predictive power of different epigenomic features across the mouse gastrulation trajectories in 3E. G Network diagram illustrating inferred pairwise interactions between transcription factor models in the mouse epiblast to mesoderm trajectory. Edge thickness represents the magnitude of the beta coefficient for each pairwise interaction. The purple edge highlights the interaction between Mesp2 and Eomes, further examined in panels H and I. H Empirical cumulative distribution function (ECDF) of dAP for loci with strong affinity for Mesp2 (red, n = 2715), Eomes (blue; n = 2844), both factors (purple; n = 136), and negative controls (gray; n = 93596), demonstrating synergistic effects in the mouse epiblast to mesoderm trajectory. Strong affinity was defined as motif energy ≥ 8. I Boxplots showing the distribution of dAP for loci grouped by the distance between Mesp2 and Eomes binding sites within the same CRE. Boxes: median (center), IQR (box), whiskers = 1.5×IQR; points outside whiskers = outliers. One-sided Kolmogorov–Smirnov tests compared the (0,20] bp group against each other spacing group. * p < 0.05, *** p < 0.001. (0,20] bp vs. others: (20,50] D = 0.1126, p = 4.94 × 10⁻²; (50,100] D = 0.1719, p = 5.99 × 10⁻⁴; (100,200] D = 0.2362, p = 3.49 × 10⁻⁷; (200,300] D = 0.4205, p = 7.47 × 10⁻⁸. J–L As in (G–I), showing data from the human HSC to CMP trajectory and the Atf4-Atf3 interaction. Sample sizes: Atf4 n = 3514; Atf3 n = 2532; both n = 251; negative controls n = 48,990. Boxplots: median (center), IQR (box), whiskers = 1.5×IQR; points outside whiskers = outliers. One-sided Kolmogorov–Smirnov tests compared the (0,20] bp group against each other spacing group. * p < 0.05, *** p < 0.001. (0,20] bp vs. others: (20,50] D = 0.0694, p = 1.49 × 10⁻¹ [ns]; (50,100] D = 0.0869, p = 4.0 × 10⁻²; (100,200] D = 0.1480, p = 5.1 × 10⁻⁵; (200,300] D = 0.1625, p = 8.1 × 10⁻⁵; (300,500] D = 0.2728, p = 2.3 × 10⁻⁹. Source data are provided as a Source Data file.