Cell-type-aware analysis of RNA-seq data

Most tissue samples are composed of different cell types. Differential expression analysis without accounting for cell-type composition cannot separate the changes due to cell-type composition or cell type-specific expression. We propose a computational framework to address these limitations: CARseq (cell-type-aware analysis of RNA-seq). CARseq employs a negative binomial distribution that appropriately models the count data from RNA-seq experiments. Simulation studies show that CARseq has substantially higher power than a linear model-based approach and it also provides more accurate estimate of the rankings of differentially expressed genes. We have applied CARseq to compare gene expression of schizophrenia/autism subjects versus controls, and identified the cell types underlying the difference and similarities of these two neuron-developmental diseases. Our results are consistent with the results from differential expression analysis using single-cell RNA-seq data. The CARseq method allows users to assess cell type-specific differential expression using RNA-seq data from bulk tissue samples, which opens up several opportunities for re-analyzing existing RNA-seq data and designing new studies.