Table 2.

Comparison between 7 clustering methods, applied on the embedding learned with contrastive-sc

	Dataset name	KMeans	Leidena	Birch	GMM	MeanShifta	Spectral Clustering	Hierarchical Clustering
1	Quake Smart seq2 Trachea	0.86	0.3	0.87	0.88	0.78	0.84	0.89
2	Quake10 × Bladder	0.75	0.29	0.75	0.75	0.72	0.75	0.75
3	Quake10 × Spleen	0.91	0.17	0.72	0.74	0.84	0.83	0.9
4	Quake Smart seq2 Diaphragm	0.98	0.7	0.98	0.98	0.96	0.98	0.98
5	Quake10 × Limb Muscle	0.99	0.58	0.9	0.91	0.69	0.96	0.99
6	Quake Smart seq2 Limb Muscle	0.98	0.67	0.97	0.92	0.92	0.97	0.98
7	Romanov	0.73	0.49	0.52	0.48	0.62	0.73	0.69
8	Adam	0.83	0.67	0.64	0.61	0.54	0.84	0.8
9	Muraro	0.91	0.57	0.85	0.85	0.83	0.85	0.92
10	Young	0.66	0.68	0.57	0.6	0.39	0.65	0.65
11	Quake Smart seq2 Lung	0.59	0.49	0.59	0.49	0.48	0.57	0.6
12	10 PBMC	0.7	0.51	0.61	0.6	0.66	0.66	0.69
13	Mouse ES cells	0.73	0.52	0.64	0.64	0.69	0.72	0.74
14	Worm neuron cell	0.49	0.49	0.34	0.36	0.15	0.47	0.49
15	Mouse bladder cell	0.46	0.56	0.53	0.5	0.51	0.42	0.44
	Average score	0.77	0.51	0.70	0.69	0.65	0.75	0.77

The results depict the average ARI score across 3 consecutive runs. The methods annotated with ^aare those where the correct number of clusters was not provided as input. The complete analysis of the remaining clustering scores has been provided in Additional file 1: Tables S6–S8

The best scores per dataset are highlighted in bold