Cell Type

How Are Cell Types Annotated in DISCO?

We performed two rounds of manual annotation for both tissue and disease atlases. In the first round, each atlas was annotated individually using marker genes from relevant publications. In the second round, we consolidated the cell types annotated across all atlases and resolved any inconsistencies in naming for the same cell types. The finalized cell types were then used as references in CELLiD, our cell type annotation tool developed in-house, and all samples in DISCO were automatically annotated using this tool.

To construct cell type-specific atlases, we extracted cells of specific cell types from samples for integration. We used FastIntegration for data integration, treating individual projects as batches to increase the number of cells per batch. Multiple rounds of integration were performed, with low-quality cells and contamination of other cell types removed at each step. Detailed annotation of cell types was then undertaken for each atlas and these cell types were subsequently incorporated into the CELLiD reference.

Cell Type Ontology

To harmonize the cell type annotation used in the atlases, we established our reference cell type ontology by integrating and harmonizing existing cell type annotation. The aim of this ontology is not only to describe the relationships between cell types but also facilitate data retrieval and downstream analysis such as cell type annotation. The first level of the ontology consists of 11 broad cell types (e.g., immune cells, epithelial cells, endothelial cells) that are manually defined. These 11 cell type groups show distinct gene expression profiles across all atlases. At subsequent levels in the hierarchy, cell types and sub-types are progressively more finely divided with assignment determined by cell-to-cell similarity and prior knowledge of cell type relationships.