Fig. 4

Benchmark results of scNPF-fusion on the Darmanis data. a Heatmaps for similarities learned from the data by Euclidean distances, pairwise Pearson correlations, SIMLR, scNPF-fusion, and RAFSIL. The scales in relative units denote the similarity. Cells with the same cell type (annotated by the colored axes) are grouped together. b t-SNE visualization for similarity matrices learned from different similarity measures. Each point denotes a cell. Smaller distance between two cells means higher similarity. True labels were not used as inputs for dimension reduction but were indicated in distinct colors to validate the results. RAFSIL1 and RAFSIL2 denote the result from the RAFSIL tool with the embedded RAFSIL1 or RAFSIL2 method