Table 1.

Benchmarking of clustering results and running time.

Data set	Methods ARI			Running time		Number of clusters
	Stable res SCORE	High res SCORE	SC3	SCORE	SC3	Stable res SCORE	High res SCORE	SC3
Baron	0.613	0.613	0.265	23.934 min	137.735 min	9	9	54
Klein	0.800	0.800	0.636	3.342 min	15.891 min	6	6	16
Camp	0.559	0.597	0.556	0.544 min	2.693 min	4	5	10
Koh	0.565	0.661	0.824	0.696 min	3.239 min	7	8	18
Kumar	0.574	1.000	0.994	0.281 min	0.496 min	2	3	4
Yan	0.588	0.588	0.650	0.108 min	0.247 min	3	3	6

Individual tests can be found through the links at: https://imb-computational-genomics-lab.github.io/scGPS/index.html. Processed data in the repository by Soneson and Robinson (2018) for the datasets generated by: Koh et al. (2016) (C1 Fludigm sequencing, 651 Human mesoderm cells, 10 Clusters) and Kumar et al. (2014) (Smart-seq, 268 mouse Embryonic stem cells, 3 Clusters). Processed data in repository from the Hemberg collection (https://hemberg-lab.github.io/scRNA.seq.datasets/) for the datasets generated by: Baron et al. (2016) (inDrop sequencing, 8569 pancreatic cells in human, 14 clusters), Klein et al. (2015) (inDrop sequencing, 2717 mouse ES cells, 4 Clusters), Camp et al. (2015) (Smart-seq, 553 human cerebral cortex organoids, 5 Clusters) and Yan et al. (2013) (Smart-seq, 90 human embryonic stem cells, 7 Clusters). Stable Res SCORE indicates the optimal resolution, most robust to parameter changes. High Res SCORE corresponds to the original number of clusters before the window-scanning procedure to search for stable clusters. The High Res SCORE corresponds to the clustering result in the first row underneath the dendrogram tree as in Figure 2A.