Skip to main content
[Preprint]. 2023 Jan 27:2023.01.26.523391. [Version 1] doi: 10.1101/2023.01.26.523391

Figure 2: Evaluation of scMINER clustering performance.

Figure 2:

a, Clustering performance of scMINER, Seurat, SC3 and Scanpy measured by adjusted Rand index (ARI). b, The average ARIs and their variance (vertical segments). scMINER significantly outperforms other clustering methods (p = 0.0004 by the one-sided Wilcoxon test). c, UMAP and silhouette plots of the Zeisel and Klein datasets using scMINER, Seurat, and SC3. Silhouette index is reported (red dashed line) for each UMAP representation of clustering result. d, Average silhouette index values and their variance (vertical lines). scMINER significantly outperforms other clustering methods (p = 0.0019 by the one-sided Wilcoxon test). e, Clustering performance comparison using four distance metrics mutual information (MI), Spearman correlation, Pearson correlation and Euclidean distance as metrics. MI outperforms other linear metrics when the number of dimensions is greater than a fixed number. f, Clustering performance comparison using four dimension reduction approaches multidimensional scaling (MDS), principal component analysis (PCA), Laplacian, and PCA sequentially followed by Laplacian (LPCA).