Separation of mixed source RNA-Seq reads by comparative genomic processing

Knowledge of the cell-autonomous and non-autonomous mechanisms operating within biological systems is essential to reveal the underlying molecular processes at work and is particularly important in functional studies of neurological diseases and cancers. These studies commonly measure gene expression levels in different cell types, but are often confounded by invasive sample processing. Indeed physical separation techniques for cell mixtures have been shown to trigger stress and apoptosis related genes, obscuring the identification of genes of interest and introducing bias. We present a novel approach that alleviates this issue for gene expression quantification using RNA-seq by conducting in silico sequence separation between different sources. Our method takes sequences from mixed species RNA-seq samples (for example two cell types each belonging to closely related species, or different strains of the same species) and differentially maps them between the two genomes. The mappings for each read are then assessed by alignment quality factors and assigned to one source genome or the other according to tuneable selection criteria. Separated read sets can then be examined using standard quantification and differential expression methods for RNA-seq data. Using simulated and real data for rat and mouse we demonstrate here that reads from closely related species can be successfully separated in this manner.