RUV-III-NB: Normalization of single cell RNA-seq Data

Abstract Despite numerous methodological advances, the normalization of single cell RNA-seq (scRNA-seq) data remains a challenging task and the performance of different methods can vary greatly across datasets. Part of the reason for this is the different kinds of unwanted variation, including library size, batch and cell cycle effects, and the association of these with the biology embodied in the cells. A normalization method that does not explicitly take into account cell biology risks removing some of the signal of interest. Furthermore, most normalization methods remove the effects of unwanted variation for the cell embedding used for clustering-based analysis but not from gene-level data typically used for differential expression (DE) analysis to identify marker genes. Here we propose RUV-III-NB, a statistical method that can be used to remove unwanted variation from both the cell embedding and gene-level counts. RUV-III-NB explicitly takes into account its potential association with biology when removing unwanted variation via the use of pseudo-replicates. The method can be used for both UMI or sequence read counts and returns adjusted counts that can be used for downstream analyses such as clustering, DE and pseudotime analyses. Using five publicly available datasets that encompass different technological platforms, kinds of biology and levels of association between biology and unwanted variation, we show that RUV-III-NB manages to remove library size and batch effects, strengthen biological signals, improve differential expression analyses, and lead to results exhibiting greater concordance with independent datasets of the same kind. The performance of RUV-III-NB is consistent across the five datasets and is not sensitive to the number of factors assumed to contribute to the unwanted variation. It also shows promise for removing other kinds of unwanted variation such as platform effects. The method is implemented as a publicly available R package available from https://github.com/limfuxing/ruvIIInb .