Fig. 2: Matching and Integration of CyTOF and CITE-seq bone marrow data.

a, t-SNE plots of individual cells colored by assay modality, either preintegration or MARIO integration. IFNG, interferon gamma. b, t-SNE plots of MARIO integrated cells colored by clustering results from (top left to bottom right): high concordance in shared cell types based on annotations from both original datasets; annotation from Levine et al.; annotation from Stuart et al. and the clustering result based on CCA scores from MARIO high cell-type resolution using information from both assays. c–e, Benchmarking results of MARIO against other mNN-based methods. c, The matching accuracy (left) and the proportion of cells being matched (right) are tested by sequentially dropping protein features. d, The matching accuracy (left) and the proportions of cells being matched (right) are measured with increasingly spiked-in noise. e, The error avoidance score (higher is better) is calculated after dropping each cell type sequentially from the dataset. f, Heatmap of cross-modality protein expression levels for the matched cells. g, t-SNE plots of the matched cells with protein or RNA expression levels overlaid based on each of the assays.