In this paper we used an automated iterative subclustering algorithm to find hundreds of cell state clusters, after which we merged unsupported clusters and annotated each resulting cluster with a better-known cell subtype. The high-res cell state clusters are annotated as subtype.gene1.gene2, so in some cases we didn't directly name some commonly named subtypes, such as pericytes. When we run DE on the resulting subtypes, we find many canonical and scRNA-suggested marker genes.
