Application of CIBERSORTx and BayesPrism to deconvolution of bulk RNA-seq data from human myocardium and skeletal muscle

Abstract

RNA-sequencing (RNA-seq) is an important tool to explore molecular mechanisms of disease. Technological advances mean this can be performed at the single-cell level, but the large sample sizes needed in clinical studies are currently prohibitively expensive and complex. Deconvolution of bulk RNA-seq offers an opportunity to bridge this gap by defining the cell lineage composition of samples. This approach is widely used in immunology studies, but currently there are no validated pipelines for researchers analysing human myocardium or skeletal muscle. Here, we describe the application and in silico validation of two pipelines to deconvolute human right atrium, left ventricle and skeletal muscle bulk RNA-seq data. Specifically, we have defined the major cell lineages of these tissues using single cell/nucleus RNA-seq data from the Heart Cell Atlas, which are then applied during deconvolution using the CIBERSORTx or BayesPrism deconvolution packages. Both pipelines gave robust estimates of the proportion of all major cell lineages in these tissues. We demonstrate their value in defining age- and sex-differences in tissue composition using bulk RNA-seq data from the GTEx consortium. Our validated pipelines can be rapidly applied by researchers working with existing or novel bulk RNA-seq of myocardium or skeletal muscle to gain novel insights.