A Universal Algorithm to Detect Rare or Novel Cell Types in High-Throughput Single-Cell Gene Expression Data

Detecting rare cell types would allow early disease detection of cancers and infections, identification of new cell types, and a deepened understanding of cell differentiation. We developed a universal algorithm to identify rare cell types from a wide variety of single-cell omics data. We validated our algorithm on single-cell qPCR data from mouse hematopoietic cells and single-cell RNA-seq data from human glioblastoma tumors cells, both with expression values from an ample number of genes. We then applied our algorithm to seq-FISH data from mouse hippocampus cells containing expression values from only 121 genes. Our algorithm detected rare cell types including a putative new hippocampal cell type.\n\nAuthor summaryRare cell type detection would advance early disease diagnosis (e.g., cancer, infection), allow identification of new cell types, and increase understanding of cell differentiation. Current computational methods can detect common cell types, but it remains a challenge to detect rare cell types within cell populations, especially with expression data from a relatively small number of genes. We created a powerful algorithm to detect rare cell types in a population of cells. We validated our algorithm on data from mouse blood stem cells and human glioblastoma tumors cells. When we applied our algorithm to mouse hippocampus data containing expression values from only 121 genes, we detected a putative new brain cell type that no previous algorithm has identified. Our universal algorithm can now be applied to a wide range of data to detect the early onset of diseases and discover new cell types.