Fig. 3

Results obtained on the PBMC datasets. a Boxplot showing the ARI values achieved by CCA, ComBat, PCA, MMDAE followed by Harmony with dimension (256, 32), PCA followed by BBKNN, and PCA followed by Harmony on the PBMC datasets. b Boxplot showing the ARI values achieved by the best AE for each of the tested dimension (H, L) of the hidden layer (H neurons) and latent space (L neurons). c UMAP visualisation of the cell-type manually annotated in the original paper. d UMAP visualisation of clusters identified by the Leiden algorithm using the resolution corresponding by the best ARI achieved by MMDAE followed by Harmony. p-value \(\le 0.0001\) (****); \(0.0001<p\)-value \(\le 0.001\) (***); \(0.001<p\)-value \(\le 0.01\) (**); \(0.01<p\)-value\(\le 0.05\) (*); p-value\(>0.05\) (ns)