Gene signatures from scRNA-seq accurately quantify mast cells in biopsies in asthma

Respiratory disease, characterized by changes in the cells of the lung, can affect molecular phenotype of cells and the intercellular interactions, resulting in a disbalance in the relative proportions of individual cell types. Understanding these changes is essential to understand the pathophysiology of lung disease. Conventional 'bulk' RNA-sequencing (RNA-seq), analyzing the entire transcriptome of the tissue sample, provides information about average expression levels of each gene in the mixed cell population; whereas it does not consider the cellular heterogeneity in samples composed of more than one cell type 1 . Single-cell RNA-seq (scRNA-seq) assesses the transcriptome of a complex biological sample with single-cell resolution, allowing identification of the relative frequency of discrete cell-types and analysis of their transcriptomes 1 . Nevertheless, analyzing the transcriptomic signature in large numbers of patients by scRNA-Seq is currently limited by its high costs. Mast cells are key regulatory cells driving the inflammatory process in asthma2 . Since they can be quantified by immunohistochemical staining for validation purposes, we used mast cells as an example of a rare cell population to assess the validity of our deconvolution approach. Recently, a number of bulk RNA-seq deconvolution methods have become available 3 , for instance of two deconvolution methods, namely support vector regression (SVR) 4 , the machine-learning method implemented in CYBERSORT, and Non-Negative Least Square (NNLS) 5 , using a matrix of cell-type selective genes identified with AutoGeneSc 6 . Both approaches are designed to estimate relative proportion of the main, common cell types present in the sample. When we used these methods to estimate the number of mast cells, we found a poor correlation with the number of mast cells stained by immunohistochemistry in the biopsies, suggesting the CIBERSORT and NNLS are less reliable in the case of rare cell types. We explored the possibility to use scRNA-Seq data from small numbers of subjects to specifically interrogate the relative cell type frequency of a rare cell population in a bulk RNA-Seq dataset obtained from a large asthma cohort