|
BOM identifies cell-type-specific CREs and cell types across species. a Binary models were trained to distinguish cardiomyocyte-specific human fetal CREs or mouse E8.25 CREs from a background set of CREs specific to other cell types. Both models were used to predict human CREs, an independent CRE set not used during training. b ROC (top) and precision‒recall (bottom) curves for the prediction of human fetal cardiomyocyte CREs via the models trained on human (blue) or mouse motif counts (pink). The area under the curve (AUC) value is shown for each patient. c Proportion of human cardiomyocyte CREs that can be aligned (gray) and cannot be aligned (black) with the mouse genome. The CREs were separated into those correctly predicted (true positives) and those incorrectly predicted as the background set (false negatives). The CREs were mapped to the mouse genome via liftOver with a threshold of minMatch ≥0.6, where minMatch represents the minimum ratio of bases that should remap. d ROC curves for the prediction of human adult heart CREs via a mouse multiclass model. e Proportion of human cells (n = 55,886 cells) predicted as each of the cell types in the mouse multiclass model. The top section of the heatmap shows the common cell types between humans and mice, and the cell types specific to the human dataset (i.e., adipocytes) are shown at the bottom. Intersections between the human cell types and the mouse model classes are highlighted in yellow squares. Source Data are provided as a Source Data file.
|
