Skip to main content
Fig. 3 | Genome Biology

Fig. 3

From: Metacell-2: a divide-and-conquer metacell algorithm for scalable scRNA-seq analysis

Fig. 3

Scaling MC2 to millions of cells. A Graphs show scaling of MC2 (multi-pile) compared to a naïve metacell on a single pile or a PCA+2-Phase Louvain clustering implementation in Seurat, using the PMBC 160K cell data (resampled to datasets of increasing sizes—X-axis). B Comparison of MC2 and two-phase clustering performance for the organogenesis datasets (MOCA). C Effects of scaling the pile sizes on the normalized inner variance for MC2 on the organogenesis data. D Distribution of normalized inner variance for MC2 and PCA+Louvain original sub-clusters on the organogenesis data. E Marker heat map and metacell graph projection of the organogenesis data. Clustering of metacells is used for coloring and cross-reference purpose, in support of, but not in place of supervised annotation. F Distribution of metacells linkage with different embryonic time points over the metacell graph. Color coding is based on metacell clustering as in D. To compensate for differences in the number of cells, we randomly sampled 2000 points for each time point and weighted by the fraction of the cells of each age in each metacells

Back to article page