To transfer cell annotations from the reference macaque striatum dataset to the human or macaque striatum cells, we performed two label transfers at increasing resolutions: one with all cells and another with just MSNs. As He, Kleyman et al. described the differences between transcriptionally and anatomically distinct MSN subtypes are subtle, so we split the annotations into two steps to optimally annotate the cells. The first label transfers the cell classes (Oligodendrocytes, MSNs, Interneurons, etc.) from the macaque to the human dataset with the Seurat functions FindTransferAnchors (reduction = ‘rpca’) and TransferData. Next, we identified cells or cell clusters that were labeled as MSNs and transferred MSN subtype labels (D1.Striosome, D2.Striomsome, etc.) from the macaque to human datasets. We filtered out cells where the cell class or cell subtype labels have max prediction scores less than 0.5 as these tend to represent noisy predictions due to low quality cells from either dataset. We confirmed accurate label transfer at the cell class and cell subtype levels with published marker genes and similar proportions across individuals and samples.
