Hierarchical classification of immune cell transcriptomes at population-scale
Journal:
bioRxiv
Published Date:
Jun 21, 2026
Abstract
Accurate immune cell classification is essential for interpreting single-cell RNA sequencing (scRNA-seq) data. However, progress in automating cell type annotation is constrained by the lack of independent, high-resolution benchmarks, as routine data integration introduces statistical dependencies that inflate model generalizability. Here, we present the single-cell universal classification omnibus (Suco), a resource of independent, uniform expert annotations, and Compocyte, a modular hierarchical classifier. Together, they establish a framework that substantially outperforms existing classifiers while facilitating expert review of ambiguous annotations. Applying Compocyte across 50 studies, including three newly generated datasets, we classified 15.6 million leukocytes from 3,965 patients. Within this cohort, we identified a new tumor-associated resorptive macrophage phenotype, a non-canonical monocyte subtype in subclinical cytokine release syndrome, and the programmatic erosion of T cell memory stemness across metastatic sites. Suco and Compocyte thus provide a generalizable framework to uncover the principles governing human immunity at population scale.
O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=101 SRC="FIGDIR/small/728980v2_ufig1.gif" ALT="Figure 1">
View larger version (37K):
[email protected]@1660b89org.highwire.dtl.DTLVardef@7f7b2corg.highwire.dtl.DTLVardef@16f9558_HPS_FORMAT_FIGEXP M_FIG C_FIG In briefThe single-cell universal classification omnibus and the modular hierarchical classifier Compocyte enable annotating single cell RNA sequencing data from 3,965 patients, revealing novel resorptive macrophage and vaccination-associated monocyte states, alongside the erosion of T cell memory stemness as a hallmark of solid tumor metastases.
HighlightsO_LISuco, a benchmark enabling novel single cell artificial intelligence models
C_LIO_LICompocyte, a hierarchical cell type classifier outperforming current architectures
C_LIO_LIMacrophages adopt osteoclast-like gene expression states across cancer types
C_LIO_LIStem-like programs erode in metastasis-infiltrating T memory cells across tumors
C_LI
