CellMATE is robust on joint profiling with different combinations and numbers of modalities. (A) UMAP embedding learned by CellMATE of scADT+scATAC (ASAP-seq), scADT+scRNA (CITE-seq), scChIP+scChIP (MulTI-tag) and scATAC+scChIP (scCUT&tag) datasets. Each dot represents a cell colored by cell type labels. (B) Clustering performance of CellMATE and the competing methods evaluated by ACC (X axis) and ARI (Y axis) on each dataset. (C-D) UMAP embeddings learned by CellMATE of the scRNA+scADT+scATAC (DOGMA-seq) dataset (C) and scChIP+scChIP+scChIP (MulTI-tag) dataset (D). Each dot represents a cell colored by cell type or lineage labels. (E) Clustering performance of CellMATE, WNN and MOFA+ on the MulTI-tag dataset, evaluated by ACC (X axis) and ARI (Y axis). (F-G) Bubble plots summarizing the performance of the methods against the best performance on each dataset by ACC (F) and ARI (G). Color represents the relative ACC or ARI score of each method against the best performance score on each dataset. NA stands where the method is not applicable.