scPANDA: PAN-Blood Data Annotator with a 10-Million Single-Cell Atlas

Recent advancements in single-cell RNA sequencing (scRNA-seq) have revolutionized the study of cellular heterogeneity, particularly within the hematological system. However, accurately annotating cell types remains challenging due to the complexity of immune cells. To address this challenge, we develop a PAN-blood single-cell Data Annotator (scPANDA), which leverages a comprehensive 10-million-cell atlas to provide precise cell type annotation.

The atlas, constructed from data collected in 16 studies, incorporated rigorous quality control, preprocessing, and integration steps to ensure a high-quality reference for annotation. scPANDA utilizes a three-layer inference approach, progressively refining cell types from broad compartments to specific clusters. Iterative clustering and harmonization processes were employed to maintain cell type purity throughout the analysis. Furthermore, the performance of scPANDA was evaluated in three external datasets.

The atlas was structured hierarchically, consisting of 16 compartments, 54 classes, 4,‍460 low-level clusters (pd_cc_cl_tfs), and 611 high-level clusters (pmid_cts). Robust performance of the tool was demonstrated in annotating diverse immune scRNA-seq datasets, analyzing immune-tumor coexisting clusters in renal cell carcinoma, and identifying conserved cell clusters across species.

scPANDA exemplifies effective reference mapping with a large-scale atlas, enhancing the accuracy and reliability of blood cell type identification.