Figure 3.

Performance comparison using different methods

A. The NMI and ARI values for each method on each dataset. Clustering was performed on 2-D representations of different algorithms and then the output was compared with true cell type labels for the 20 datasets indicated. Detailed dataset information is listed in Table 1. B. The statistics of the ranks of the compared methods based on NMI and ARI values. For each dataset, NMI and ARI values given by different algorithms were ranked in the descending order, with rank 1 indicative the highest NMI or ARI values. The number of ranks achieved by these algorithms in the 20 datasets is then counted for distribution. NMI, normalized mutual information; ARI, adjusted rand index.