Fig. 3

Clustering results. a The adjusted Rand Index (ARI) plotted against various number of clusters; each subpanel represents a dataset. Each point represents a clustering solution obtained by varying the resolution parameter and the random seed in Leiden algorithm. The line plot is the average ARI at a given number of clusters. b Heatmaps displaying the normalized areas under the curve (AUC) from plots similar to a, but for various partition-level metrics. The color scale indicates deviations from the column-wise median scaled by the matrix-wise median absolute deviation [23]. It provides a comparison of relative performance between methods, and is unified across datasets and robust to outliers