Extended Data Fig. 3: Iterative clustering and quality control.

a, The expression of markers in first round of clustering (solid lines) and second round of clustering (dashed lines) in brain. Markers that support second round of clustering were shown. Two color bars on top denote clustering results of first round and second round, respectively. b, The expression of markers relative to boundaries of one round of clustering and second round of clustering in endothelium. Convention follows panel a. c, The number of clusters resulted from a series of resolution ‘r’ and PCs in the clustering of spinal neuron. d, The pairwise ARI between clusters resulted from different resolution ‘r’ in spinal neuron. e, The mean ARI of clusters in testing resolution ‘r’ and PCs (Methods). Each dot denotes a super-cluster. f, Cross-validation on clustering by scPred²¹. The first column shows the AUROC (area under receiver operating curve) of testing the identity of developmental systems (each red dot is a system). Other columns show the AUROC of testing ‘Celltype_annotation’ and ‘Final_annotation’ (Supplementary Table 1) within each system (expect PGC) in red and blue dots, respective (Methods, n = 3~59 cell types). Testing of randomly shuffled identity was served as control in each column (grey). The center line denotes the median, while the box contains the 25th to 75th percentiles. The whiskers mark 1.5x interquartile range. g, Batch effects of embryos, technical replicates, cell cycle phase, and total UMIs estimated by the entropy of mixing¹⁰⁰ (Methods, n = 1,700 randomly sampled cells). ‘+Ctrl’, cluster identities as batch. ‘-Ctrl’, randomly assigned batch. Boxplots are defined as in panel f. h, UMAP of all cells colored by embryos, technical replicates, cell cycle phase, total UMIs, and developmental systems in each embryo. The missing parts in embryos 03, 06, and 07 are due to 4 libraries that did not pass quality control (Methods).