STSM on advancing in silico myeloid cell deconvolution from bulk transcriptomics
Figure: © Springer Nature, Finotello et al.
Bulk tissue samples are composed of different cell types at varying fractions which can be inferred using techniques such as fluorescence-activated cell sorting, which is prohibitively laborious and expensive for clinical use. Since these fractions are informative of the interaction of the immune system with diseases, in silico cell-type deconvolution can be considered as an alternative to infer cell type fractions from bulk samples based on cell-type specific gene signatures. First-generation deconvolution methods such as CIBERSORT or quanTIeq (bundled in immunedeconv) are already able to distinguish different types of immune cells including myeloid cells, but previous work has shown that myeloid cell types are not well quantified by these methods with, for instance, issues separating monocytes and macrophages, as well as dendritic cell subtypes. Second-generation methods on the other hand, infer suitable signatures from single-cell RNA-seq data and are hence more flexible. However, it is currently unclear how robustly these methods can separate myeloid cell types in different data sets and applications. The purpose of this STSM was to explore different second generation and unsupervised deconvolution methods on a range of single-cell data sets to assess the robustness of gene signatures, with the ultimate goal of establishing best practices for the deconvolution of myeloid cells.
Figure citation link: https://rdcu.be/c7Lay