PRECISION AND ACCURACY IN SINGLE-CELL RNA-SEQ

Abstract

Single-cell/nuclei RNA sequencing (sc/snRNA-seq) is widely used for profiling cell-type gene expression in brain research. An important but frequently underappreciated issue is the data quality in terms of precision and accuracy. We evaluated precision using data from 14 human brain studies with a total of 3,483,905 cells from 297 individuals, with technical replicates based on random grouping of cells of the same type from the same individual. We also evaluated accuracy with sample-matched scRNA-seq and pooled-cell RNA-seq data of cultured mononuclear phagocytes from four species. Low precision and accuracy at the single-cell level across all evaluated data were observed. Cell number was highlighted as a key factor determining the expression precision, accuracy, and reproducibility of differential expression analysis in sc/snRNA-seq. A high missing rate is likely the cause of the quantification quality problem. Downstream analysis results are severely affected by the expression quality issue. Many false findings can be produced when the noises are not properly controlled. This study underscores the necessity of sequencing enough cells per cell type per individual, preferably in the hundreds, to mitigate noise in expression quantification. Pseudo-bulk aggregation of expression data over cells of the same type is required when the high-quality expression quantification is desired.