Next-generation deconvolution of the tumor microenvironment with omnideconv.

Abstract

The tumor microenvironment and, particularly, tumor-infiltrating immune cells can profoundly influence tumor progression and response to therapy. Deconvolution is a powerful computational technique to estimate cell-type fractions from bulk RNA sequencing (RNA-seq) data leveraging expression signatures specific to the cell types of interest. Recently, a new generation of deconvolution algorithms has emerged, making it possible to directly learn cell-type-specific signatures to be used for deconvolution from annotated single-cell RNA-seq (scRNA-seq) datasets. Thanks to their flexibility, these next-generation methods can extend deconvolution to any cell type, tissue, and organism for which a suitable single-cell reference is available. However, these methodologies are highly diverse in terms of programming languages, computational workflows, and input/output data, which complicate their usage and comparison. To overcome these challenges, we developed omnideconv, an R package that integrates several deconvolution methods, streamlining their usage and unifying their semantics. In this chapter, we demonstrate how omnideconv can be integrated with an annotated scRNA-seq dataset, comprising both malignant and normal cells from the breast cancer microenvironment, to quantify the cellular composition of bulk RNA-seq data from a cohort of breast cancer patients.