Fig. 1

Overview of MultiK workflow. a Step 1: Starting from a count matrix, MultiK subsamples 80% of the cells and performs a standard scRNA-seq pipeline using Seurat (version 3.1.2) 100 times over each of 40 resolution parameters (from 0.05 to 2 with an increment of 0.05). Then, MultiK determines candidate Ks based on the frequency of K and rPAC across all subsampling runs. b Step 2: MultiK assigns class or subclass labels to clusters at candidate K levels. MultiK first constructs a dendrogram of the cluster centroids using hierarchical clustering and then runs SigClust on each pair of terminal clusters to determine classes and subclasses