More cells, more doublets in highly multiplexed single-cell data

Sample barcoding allows deconvolution of multiplets in multiplexed droplet-based single-cell RNA-sequencing experiments. However, this is only possible when each cell comes from a different sample. As the number of cells in a droplet increases, the probability of two or more cells coming from the same sample increases rapidly. We show that the number of these unresolvable multiplets is greater than previously estimated for the 10X Flex scRNA-seq protocol, and provide a formula for estimating the fraction of multiplets in a data set given a measured average droplet occupancy and number of unique samples in a pool. We also show that existing doublet detection tools should be applied to Flex data to identify these multiplets, and demonstrate that filtering out barcodes identified by these tools improves downstream analysis.