Identifying cell states in single-cell RNA-seq data at statistically maximal resolution
Fig 2
(A) 2D-histogram of homogeneity and completeness of inferred cellstate memberships in 162 simulated datasets. The inset shows the distributions in the region [0.95, 1], [0.95, 1] where all results fall. (B) Comparison of the performance of Cellstates with those of other clustering tools. For Cellstates, we show the results for the full partition into cellstate-groups (“cellstates”) and merged to the same number of clusters as the annotation (“cellstates_hierarchical”). For SuperCell we similarly show results with fine-grained setting matching the cluster number of Cellstates and with coarse-grained setting matching the cluster number of the annotation. In each plot, we show homogeneity and completeness of the partitions obtained by the different methods using the published annotation as the reference partitions. Note that low homogeneity and completeness may indicate partitions with, respectively, too few and too many clusters compared to the reference partition.