Hierarchical classification of immune cell transcriptomes at population-scale.

Area of Science:

• Immunology
• Computational Biology
• Genomics

Background:

• Accurate immune cell classification is crucial for single-cell RNA sequencing (scRNA-seq) data analysis.
• Existing methods lack independent, high-resolution benchmarks, leading to inflated generalizability due to dataset integration.
• Progress in immune cell classification is hindered by the absence of robust validation standards.

Purpose of the Study:

• To establish a comprehensive resource of independent, expert annotations for immune cell classification.
• To develop a novel, modular hierarchical classifier for large-scale immune cell annotation.
• To create a generalizable framework for high-resolution immune cell annotation in human population immunology.

Main Methods:

• Development of the single-cell universal classification omnibus (Suco), a curated dataset of expert annotations.
• Creation of Compocyte, a modular hierarchical classifier designed for scalability.
• Application of Compocyte across 50 studies, analyzing 15.6 million leukocytes from 3,965 patients.

Main Results:

• Suco and Compocyte demonstrate superior performance compared to existing immune cell classifiers.
• Identification of a novel tumor-associated resorptive macrophage phenotype.
• Discovery of a non-canonical monocyte subtype in subclinical cytokine release syndrome and characterization of T cell memory erosion in metastatic cancer.

Conclusions:

• Suco and Compocyte provide a robust, generalizable framework and benchmark for high-resolution immune cell annotation.
• This approach facilitates expert review and enables analysis at the scale of human population immunology.
• The findings offer new insights into immune cell heterogeneity in various clinical contexts, including cancer and inflammatory conditions.