Fig. 3: Benchmarking of MaxFuse versus other integration methods across multiple ground-truth data types.
From: Integration of spatial and single-cell data across modalities with weakly linked features
a, Four different multiome datasets, generated by different technologies, were benchmarked. Cell-type matching accuracy, FOSCTTM, FOSKNN (with k = 0.5% total cell counts of each dataset), and ARI and Silhouette F1 were evaluated across all five methods. b, UMAP visualization of MaxFuse integration results for the four ground-truth multiome datasets, colored by modality (top panel) and cell type (bottom panel). BM, bone marrow; DC, dendritic cells; EMP, erythro-myeloid progenitors; mem, memory; prog, progenitor; trans, transitional.
