Identify Neuron Cells from Head Tissue at Single-Cell Level

Single-cell RNA sequencing (scRNAseq) technology has recently developed rapidly to better understand biological processes and cellular functions. As an important step in analyzing scRNAseq data, cell typing has attracted a significant research effort. Most cell typing algorithms currently cluster cells by shared nearest neighbors and then evaluate the affinity of each cluster to a type of cell based on pre-defined gene markers from prior knowledge such as CellMatch, CellMarker and PanglaoDB databases. Cell assignment was selected with the highest value of area under the curve (AUC) cross all cell-type affinity evaluation. The accuracy of the clustering-based cell typing method is affected by the resolution of the clustering and the specificity of the marker genes. In addition, assigning cell type to a cell subpopulation not individual cells loses the advantage of single-cell technology. To address this problem, we use a non-clustering method for cell typing, which no longer assigns cell types to cell subpopulations, but directly assigns cell types to each cell. This method uses multiple correspondence analysis (MCA) to calculate gene to cell distance in Barycentric coordinates, evaluates gene-cell affinity through expression levels and the number of marker genes, and combines these data to determine cell type association for each cell. Both cluster-based and MCA methods have been applied to identify neuron cells from mice head tissue. The cluster-based method has led to 7 clusters and 6 cell types while the MCA method identified 14 cell types and one unassigned group. Two cell types identified with cluster-based methods were also found with the MCA method. In MCA-based non-clustering-based method, cell typing with enrichment of marker genes with hypergeometric test and has better accuracy.