Extended Data Fig. 5: Machine learning model predicts features of buffered and linearly sensitive enhancers.
From: Precise modulation of BRG1 levels reveals features of mSWI/SNF dosage sensitivity
a, Illustration of ChromBPNet predict chromatin accessibilities at Psmd7 gene locus. Observed track: signals detected by ATAC-seq experiments; Predicted track: ChromBPNet predicted signals; Contribution track: the genome sequence scaled according to ChromBPNet contribute scores. b, Scatter plot showing the correlation between predicted and observed signals. c, Motifs identified by ChromBPNet models in G1 enhancers (left), and G5 enhancers (right). d, Marginal footprints of TF motifs of different groups using predicted profiles from mESC ATAC-seq ChromBPNet models. e, Cumulative percentage of GC content in different groups. P values were calculated with Wilcoxon’s t-test (two-sided) by comparing each group with G1: p = 2.5 × 10−4 (G2); p = 4.2 × 10−8 (G3), p = 1.3 × 10−10 (G4), p = 1.1 × 10−14 (G5). *** p < 0.001 f, Histograms showing the number of ChromBPNet predicted high score sites from the 0 nM mESCs model or 100 nM dTAG13 mESCs model. g, Representative genome browser view of high-score site at buffered or linear enhancers.
