Investigating Sources of Zeros in 10× Single-Cell RNAseq Data

Single-cell RNA sequencing allows expression profiling of hundreds of thousands of individual cells in a single experiment. The main drawback is that on the single-cell level observed proportion of zero counts is much higher than on the bulk level. In this study, we performed the analysis of potential sources of excessive zeros using multi-omics data from a homogenous breast cancer cell line. A comparison of the expression data at the population and single-cell level showed that variability between sequencing platforms is higher than when comparing replicates on the same platform. The non-linear model was used to estimate the difference in the expected and observed number of zeros per gene. Then, using gene set enrichment analysis, we discovered some biological pathways containing genes with an increased or reduced number of zeros, like ribosomal genes. Finally, we analyzed different technical factors potentially influencing the dropout rate, and found that the number of transcripts per gene, low mappability and difference in transcript coverage uniformity might cause fluctuations in gene expression estimate on a single-cell level.