We trained a support-vector machine classifier to automatically assign each cell to one of seven major classes: neurons, oligodendrocytes (all ∼236,000), astrocytes, ependymal cells, peripheral glia (e.g., Schwann cells, satellite, and enteric glia), immune cells, and vascular cells (Figure S1B). Neurons were most prevalent in rostral regions of the CNS, as well as in the cerebellum. In caudal regions, oligodendrocytes—needed to support long-range neurotransmission—dominated greatly, comprising 84% of cells in the hindbrain (excepting cerebellum) and 71% in the spinal cord. Astrocytes ranged from 13% of cells in the telencephalon to 6% in the hindbrain. Due to sources of bias, such as differential survival or cell capture, these estimates can only be approximate. As further validation, we compared with estimates obtained by single-molecule mRNA fluorescence in situ hybridization (osmFISH; Codeluppi et al., 2018) from the somatosensory cortex (Figure S1G). Notably, interneurons were undersampled by scRNA-seq, likely due to their fragility, whereas astrocytes were undersampled by osmFISH due to the difficulty of image segmentation for irregularly shaped cells with small somata.
