NLM DIR Seminar Schedule

Abstract:

The development of single-cell sequencing technologies has significantly enhanced our understanding of cell populations in complex tissues. This increase in sequencing data has led to the creation of a series of tissue-specific single-cell atlases, providing comprehensive resources for exploring cellular diversity and function. Identifying reliable marker genes is essential for accurately classifying cell types in these atlases, acting as signatures for specific cell identities. To ensure precision, cell-type markers must undergo thorough evaluation and validation, including assessments of their specificity, sensitivity, and reproducibility across various tissues. For the systematic evaluation of marker gene identification approaches, we used a single-cell tissue atlas of the human retina that includes a broad range of cell types, such as bipolar cells (BCs), retinal ganglion cells (RGCs), retinal pigment epithelium (RPE), amacrine cells, and their granular subtypes. For this evaluation, we created different combinations of marker gene sets, including local-markers, global-markers, class-markers, and combined-markers, to assess their specificity and sensitivity across cell types. Global-markers are derived from the entire dataset that includes all cell types (global dataset). Local-markers are generated from a subset of data containing only cells and cell types within a broader cell class (local dataset). Class-markers combines granular cell types within a broader category into a single cell class within the global dataset for marker gene selection. NS-Forest v4.1 was used to generate and evaluate these markers across all retinal cell types. Our results showed that local markers don’t necessarily show optimal performance globally, and global markers don’t necessarily show optimal performance locally. However, by combining local-markers with the class-markers (combined-marker, tested against the global dataset), both local and global performance can be optimized. Thus, we propose an alternative method for identifying marker sets: combining class-markers and local-markers when a robust set of global markers for the cell type is unavailable.