Exploring Hierarchical Structures of Cell Types in scRNA-seq Data

The development of single-cell transcriptome sequencing technology has provided a deeper understanding of cell types. Cell types consist of hierarchical structures that perform functions differently. Constructing a hierarchical structure of cell types is crucial for revealing relationships between cell types. Existing hierarchical methods construct cell type hierarchy with fixed branches which cannot reflect actual cell type hierarchy, and cannot adapt to large-scale unlabeled data. To fill the gap, we propose scHSD, a hierarchical structure detecting method which allows construct cell type hierarchy that closely actual structures, and is capable of handling large-scale data. scHSD first constructs a cell-cell similarity graph and conducts coarse-graining to minimize graph size. Then, scHSD leverages structural entropy minimization on the coarse-grained graph to build cell type hierarchy. scHSD adjusts hierarchy height and the number of branches during structural entropy minimization. To validate the method’s effectiveness, we conduct experiments on datasets of different scales and compared the hierarchical structure with known cell type hierarchy. The experimental results demonstrate that our method accurately construct cell type hierarchy and can be used for cell clustering and classification. This study provides a new perspective for analyzing cell types in scRNA-seq data, contributing to a deeper understanding of the relationships between cell types and cellular heterogeneity.