Skip to main content
. Author manuscript; available in PMC: 2021 Apr 1.
Published in final edited form as: Nat Mach Intell. 2020 Oct 5;2(10):607–618. doi: 10.1038/s42256-020-00233-7

Figure 2.

Figure 2.

Comparison of ItClust with unsupervised methods on human pancreatic islet datasets. (a) The boxplots of clustering ARI of ItClust, Louvain2, DESC13, and SAVER-X14 for the four human pancreatic islet datasets and the combined human pancreatic islet dataset15–18. Boxplots show the median (center line), interquartile range (hinges), and 1.5 times the interquartile range (whiskers). Resolution of Louvain, DESC, and SAVER-X ranged from 0.2 to 2.0 with a step of 0.2. (b) The t-SNE plots for the combined human pancreatic islet dataset. The coordinates of the three plots in the first row are based on ItClust clustering result and colored by ItClust clusters, true cell types, and technical batches, respectively. The coordinates of the three plots in the second row are based on Louvain clustering result with resolution set to 2.0 and colored by Louvain clusters, true cell types, and technical batches, respectively. The optimal resolution is the resolution that had the highest silhouette coefficient across a range of resolutions from 0.2 to 2.0. (c) The Sankey plots of ItClust, Louvain, DESC, and SAVER-X clustering results for the combined human pancreatic islet dataset. Resolution of Louvain, DESC, and SAVER-X was set to the optimal resolution.