Fig. 1. RP mode reveals two distinct TF classes: short-range and long-range.

a Schematic of the regulatory potential (RP) model. The regulatory effect of TF i on gene j is modeled as the RP, Ri, j(Δ), which sums up all TF i ChIP-seq binding effects on the gene j. The effect of a single binding site k of TF i on gene j decays exponentially with increasing $x_{ijk}$, the genomic distance between TSS of gene $j$ and TF $i$ binding site $k$. The exponential decay function ($2^{\frac{-x_{ijk}}{\mathrm{\Delta }}}$) is parameterized by the decay distance (Δ), the distance at which the TF regulatory effects are halved. b TF $i$-specific regulatory decay distances (${\mathrm{\Delta }}_i^{*}$) can be inferred as the Δ that best separates TF $i$ perturbation-induced differentially expressed (DE) genes from other genes. $R_{i,j}\left(\mathrm{\Delta }\right)$ with short-range (<1 kb) best separates FOXM1-knockdown or GABPA-knockdown DE gene sets (left). AR overexpression or ESR1-knockdown DE gene sets are best separated by $R_{i,j}\left(\mathrm{\Delta }\right)$ with long-range Δ (>10 kb). The two-sided Kolmogorov–Smirnov two-sample test is used to estimate the degree of separation of DE genes from other genes. c ${\mathrm{\Delta }}_i^{*}$ can also be inferred as the $\mathrm{\Delta }$ that leads to the best concordance between TF $i$ regulatory effects estimated by TF $i$ ChIP-seq ($R_{i,j}\left(\mathrm{\Delta }\right)$) and expression cohorts (${\rho }_{i,j}^{\mathrm{expr}}$: TF i-gene $j$ expression correlations), respectively. A second correlation coefficient ${\rho }_i^{\mathrm{expr},\mathrm{RP(}\mathrm{\Delta }\mathrm{)}}$ was calculated to measure the concordance between ${\rho }_{i,j}^{\mathrm{expr}}$ and $R_{i,j}\left(\mathrm{\Delta }\right)$ (see the main text for the rationale and Methods for statistical details). d TFs with short-range ${\mathrm{\Delta }}_i^{*}$ (100bp-3 kb) include YY1, CREB1, FOXM1, ATF1, and TFDP1 (left). TFs with long-range ${\mathrm{\Delta }}_i^{*}$ (3 kb–100 kb) include PPARG, FOXA1, GRHL2, FOSL2, and TEAD1 (right). Colored shaded regions depict the 95% confidence intervals derived from all ChIP-seq samples that passed QC for each TF. Dots along the line are Δ values being tried. e Distribution of regulatory decay distances (${\mathrm{\Delta }}_i^{*}$) of 11 short-range TFs (left) and 49 long-range TFs (right). Source data are provided as a Source Data file.