Fig. 6

The predicted H3K4me3 signal and corresponding attributions for two cell types in the same region of chromosome 20. a The prediction and attributions for GM12878, where a tall peak on the right is paired with two much smaller peaks to the left. Many short regions have a positive genomic attribution but a negative cell type attribution that masks them. b The prediction and attributions for the duodenum smooth muscle. A prominent peak is now predicted on the left, corresponding with a swap from a negative cell type attribution to a positive one. The same short regions that previously were masked by the cell type attributions now have positive cell type attributions and exhibit peaks in the imputed signal