After an initial quality assessment of samples and cells, we retained 492,949 cells as inputs to the computational analysis. During three stages of manifold learning and clustering, we removed additional doublets, outliers, and low-quality cells (Figure S1A). As oligodendrocytes are extremely abundant in the hindbrain and spinal cord, we removed more than 200,000 oligodendrocytes from these regions in order to better balance the number of oligodendrocytes between tissues (but analyzing the full set of cells did not reveal any additional structure in the oligodendrocyte lineage). The final, high-quality curated compendium comprised 265 clusters represented by 160,796 high-quality single-cell transcriptomes (Figures 1B and 1C). This represents a highly conservative clustering, and significant heterogeneity likely remains within many of the reported clusters.
