Fig. 2
From: Contrastive self-supervised clustering of scRNA-seq data
Simulated data analysis. A set of 12 simulated balanced (a–c) and 12 imbalanced (d–f) datasets has been analyzed. For simplicity, only one external (ARI) and one internal (Silhouette) evaluator across all datasets are displayed. The complete analysis is provided in Additional file 1: Fig. S1, S2. Each method processed each dataset 3 times with different initialization seeds. The error as the relative difference between the predicted and the true number of clusters [(pred − true)/true] is illustrated in c for balanced data and f for imbalanced data. The methods annotated with (*) are those that did not receive as input the number of clusters. Most methods in this category tend to overestimate the number of clusters in the data, behavior which is more pronounced in the imbalanced setting