Fig. 3
From: Contrastive self-supervised clustering of scRNA-seq data
Real scRNA-seq data analysis. A method comparison has been performed across all 15 real datasets in terms of ARI—Adjusted Rand Index (a), NMI—Normalized Mutual Information (b), Silhouette (c), Calinski (d) scores, the execution time (e) and the error of the predicted number of clusters (f). The execution time and the Calinski scores have been depicted in log scale. The error as the relative difference between the predicted and the true number of clusters [(pred − true)/true)] is illustrated in f. The methods annotated with (*) indicate those where the clustering was performed without inputting the number of clusters. All methods in this category tend to overestimate the number of clusters in the data and do more so in the imbalanced setting. For each dataset and method, three runs have been performed