SuperCell2.0 enables semi-supervised construction of multimodal metacell atlases

Multimodal single-cell atlases comprising hundreds of thousands of cells enable the exploration of complex tissues and the generation of testable hypotheses. To facilitate the analysis of such large datasets, we developed SuperCell2.0, a robust workflow for constructing (semi-)supervised multimodal metacells. Using this framework, we built multimodal metacell atlases from blood (PBMC CITE-seq) and the tumor immune and stromal microenvironment (TISME 10x Multiome). Analysis of these atlases identified interferon-primed monocytes and macrophages in circulation and in tumors. Metacell-derived markers enabled sorting and phenotypic characterization of this population in healthy donors using FACS and bulk RNA-seq. This Zenodo deposit includes: Cell annotations for both multimodal metacell atlases (one row per cell, including metacell assignment and annotation levels) PBMC_CITE_seq_sc_annotations.csv (final annotation: refined_celltype) TISME_multiome_sc_annotations.csv (final annotation: fine_celltype) Salmon transcript- and gene-level estimated count matrices for CD14 monocyte bulk RNA-seq (generated using nf-core/RNA-seq) Containers used in the study: supercell_multiomics_v2.sif (metacell benchmarking and atlas construction/analysis) edger_v1.sif (differential expression analysis) rnaenv_v1.sif (bulk RNA-seq analysis) workflow_additional_data, containing required files to reproduce the main analyses using our workflows on GitHub: SuperCellMultiomicsHTAN SuperCellMultiomicsAnalysis