Assessing the reliability of spike-in normalization for analyses of single-cell RNA sequencing data.

Abstract

By pro ling the transcriptomes of individual cells, single-cell RNA sequencing provides unparalleled resolution to study cellular heterogeneity. However, this comes at the cost of high technical noise, including cell-spec c biases in capture efficiency and library generation. One strategy for removing these biases is to add a constant amount of spikein RNA to each cell, and to scale the observed expression values so that the coverage of spike-in transcripts is constant across cells. This approach has previously been criticized as its accuracy depends on the precise addition of spike-in RNA to each sample. Here, we perform mixture experiments using two di erent sets of spike-in RNA to quantify the variance in the amount of spike-in RNA added to each well in a plate-based protocol. We also obtain an upper bound on the variance due to di erences in behaviour between the two spike-in sets. We demonstrate that both factors are small contributors to the total technical variance and have only minor e ects on downstream analyses such as detection of highly variable genes and clustering. Our results suggest that scaling normalization using spike-in transcripts is reliable enough for routine use in single-cell RNA sequencing data analyses.