Fig. 2

A The SCIPIO-86 dataset showing 288 scRNA-seq clustering pipelines applied to 86 datasets in terms of 4 unsupervised metrics (CH, DB, SIL, GSEA). Pipeline performance can be seen to have a monotonic relationship with clustering resolution for the CH, DB, and SIL metrics, with additional dependency on different pipeline strategies (filtering, normalization) also evident. B After correction for number of clusters, we embedded each scRNA-seq clustering pipeline into UMAP space, with variation visible due to normalization strategy