Droplet scRNA-seq is not zero-inflated

Abstract

Potential users of single-cell RNA-sequencing (scRNA-seq) often encounter a choice between high-throughput droplet-based methods and high-sensitivity plate-based methods. There is a widespread belief that scRNA-seq will often fail to generate measurements for some genes from some cells owing to technical molecular inefficiencies. It is believed that this causes data to have an overabundance of zero values compared to what is expected from random sampling and that this effect is particularly pronounced in droplet-based methods. Here I present an investigation of published data for technical controls in droplet-based scRNA-seq experiments that demonstrates that the number of zero values in the data is consistent with common distributional models of molecule sampling counts. Thus, any additional zero values in biological data likely result from biological variation or may reflect variation in gene abundance among cell types or cell states.