Fig. 1.

Study design. Seven independent datasets were collected, pre-processed, and clustered using the Seurat package. Cell-level metadata on cell type classification and sample clinical information was standardized for all datasets. To obtain a large reference dataset, five datasets were integrated in an anchor-based manner. The cells of the integrated reference were subjected to a standard workflow for clustering and cell type annotation. The integrated reference was used for annotation of the validation dataset and the two datasets were then merged into the final dataset. Additionally, pseudotime trajectory analysis of selected clusters was conducted.