Fig. 4.

 Accuracy evaluation of the inferred number of clusters. (a–e) Correlations between inferred cluster numbers from SC3, CIDR, Seurat, t-SNE + k-means and SAFE-clustering, respectively, and the true cluster numbers, across the 12 benchmarking datasets. (f) Average deviations between the inferred and the true numbers of clusters, measured by $\stackrel{-}{D}=\frac{1}{m}{\sum }_m|\widehat{k}-k_t|$, where the number of datasets $m$ equals to 12