Fig. 6

Similarity of partitions for multiple single-cell preprocessing parameters. A Each node represents a putative cell-type partition output by SE2, derived from unique preprocessing parameters (see legend). Edges between partitions show similarity of partitions with given preprocessing parameters, averaged over seven datasets. B Low (blue), medium (gray), and high (red) ARI values with gold-standard cell types for each dataset visualized. Edge thickness is proportional to partition similarity. C Classic/”Newman’s” modularity (Q) for each cell–cell similarity matrix derived from the SE2 partition of a dataset generated with a particular set of preprocessing parameters. D Modularity density (Qds) for each cell–cell similarity matrix derived from the SE2 partition of a dataset generated with a particular set of preprocessing parameters. E For a single brain-based single-cell dataset [70], SE2 is applied to data preprocessed with 144 distinct settings. F ARI with ground truth cell types. G Classic/”Newman’s” modularity (Q) of input adjacency matrix of cell similarities. H Modularity density (Qds) of input adjacency matrix of cells