Fig. 5

BOM provides local and global motif importance scores.

a Genome browser tracks showing the snATAC-seq signal around a region of mouse chromosome 17 near the Nkx2-5 gene for each mouse E8.25 cell type. Locations of the three cardiomyocyte-specific CREs are shown at the bottom. b SHAP local explanation of three cardiomyocyte-specific CREs shown in (a). The top four most important motifs for classifying those CREs are shown. The red and blue arrows indicate the sign (and direction) of the SHAP values. A representative name was given to every motif, and the motif count is indicated. c Heatmaps representing the predicted probability of the CREs shown in (a and b) by each of the binary models trained to predict CREs specific to a cell type in mouse E8.25. d Mean |SHAP values| of the top TF binding motifs in distinguishing cardiomyocyte-, endothelium-, and neural crest-associated CREs in mouse E8.25. The SHAP values were calculated based on the sets of CREs specific to other cell types. e Mean SHAP values for the TF binding motifs are shown in (d). f Mean expression of the TFs that bind to the motifs in (d) and (e). Expression data are normalized counts from matched scRNA-seq experiments⁵⁴. g Top 20 most important motifs in distinguishing mouse E8.25 endothelium CREs. Motifs are ranked by the mean of the absolute SHAP values. Bars colored yellow or gray, represent positive and negative SHAP values, respectively. h Normalized motif counts of the motifs shown in (g). The motif counts normalized as (d). The CREs shown are from the dataset used for endothelium CRE classification, where the background set is composed of CREs specific to the other 16 cell types in mouse E8.25 (n = 1760, 1768; background and endothelium CREs, respectively). Motifs were hierarchically clustered. Cis-BP 2.0 motifs for M. musculus from the MEME Suite were used in this figure for direct comparisons between motifs and TFs⁵⁵. Source Data are provided as a Source Data file.