Abstract:
|
Recent advances in the single cell RNA sequencing (scRNA-seq) technology have enabled investigators to assess transcriptome-wide differences at the single cell resolution. Because of the high heterogeneity in environmental exposures, immune responses, and genetic background across different subjects, subject contributes to a major source of variation in scRNA-seq data which severely confounds with disease/phenotype effect in the analysis to identify disease/phenotype associated gene signatures. Current studies either treat cells from the same subject as independent or take the average across all cells from the same subject so that methods developed for bulk RNA-seq data can be applied. To address this limitation, we developed a mixture mixed effects model for gene differential analysis. The model accounts for both dropout events and subject effects. Permutation studies using real scRNA-seq datasets showed a highly inflated type I error for methods that ignore subject effects or average across cells from the same subject. In contrast, our method has better control of type I error and is more powerful to detect differential gene expression in scRNA-seq data with multiple subjects.
|