Fig. 6: Predicting cardiac cell types, anatomical zones, and laterality.

a Diagram of hierarchal random forest model. Train and test data are aligned via scaling. Train data is used to derive a random forest model for cell type, which is then applied to test data. For cardiomyocytes the procedure is iterated for predicting anatomical zone and laterality. After prediction, anomaly detection removes cells types from the test dataset that were not present in the training dataset. b Sankey diagram of 10-fold cross-validation results on data from Cui et al.²⁹. Table provides cell type, conditional zone, and conditional laterality/side accuracies. c Sankey diagram and accuracies of prediction results for training on the Cui et al. data and prediction on the Asp et al data³⁰. d Sankey diagram and conditional accuracies of prediction results on Miao et al. data²³ (training again on the Cui et al. data). cm cardiomyocytes, ep epicardial cell, f fibroblast, ec_vasc vascular endothelial cells, ec_endo endocardial endothelial cells, v valvar, mf macrophage, b/t b/t cell, 5w undifferentiated cells, cm_myzo Myoz2-enriched CMs, f-lv fibroblast-like: (related to larger vascular development), f-sv fibroblast-like: (related to smaller vascular development), sm/f smooth muscle cells/fibroblast-like), f-skl fibroblast-like: (related to cardiac skeleton connective tissue), ep-der epicardium-derived cells, ec-p-a endothelium/pericytes/adventitia, im immune cells, eryth erythrocytes, peri pericyte, ns nervous system, LA left atrium, RA right atrium, LA left ventricle, RV right ventricle.