Computational immune cell deconvolution methods utilizing gene expression profiling to estimate immune cells in bulk tissues and blood are an appealing alternative to flow cytometry.
The main focus of this study is a comprehensive comparison between immune cell deconvolution signatures generated by xCell and Cibersort and corresponding flow cytometry data.
Gene expression and flow cytometry data from the publicly available rheumatoid arthritis GSE93777 dataset were used for validation of CIBERSORT and xCell immune cell deconvolution methods. All GSE93777 samples were used in the validation (rheumatoid arthritis patients with or without drug treatment and healthy volunteers). Treated rheumatoid arthritis patients received either methotrexate, infliximab, or tocilizumab in roughly equal parts. CIBERSORT and xCell deconvolution cell signatures were matched to corresponding/related immune cell subtypes assessed by flow cytometry. Flow cytometry cell types were manually matched to the available immune cell deconvolution signatures.
https://cibersort.stanford.edu/ CIBERSORT deconvolution algorithm provides quantification for 22 immune cell subtypes. It is a widely used tool that requires an input matrix of reference gene expression signatures, collectively used to estimate the relative proportions of each cell type of interest. To deconvolve the mixture, a linear support vector regression (SVR), machine-learning approach is utilized and the resulting immune cell abundance estimates the fraction of immune cell types in a sample. The term fraction is defined for each sample as the sum of values across the 22 cell types with a total of 1, and should reflect the true fraction of cells in the sample of a given type. In this analysis, the CIBERSORT deconvolution method was performed on all 22 immune cell subtypes.
It is a gene signatures-based method that is established on single sample gene set enrichment analysis (ssGSEA) and estimates the scores of 64 immune cell type. It is based on 489 geneset signatures extracted from large-scale expression data from different projects and studies: FANTOM, BluePrint, ENCODE, IRIS, HPCA and Noverstern (Aran D, et al. Genome Biol. 2017). It also employs a compensation technique to reduce spill-over effects between closely related cell types. Although the final xCell scores cannot be directly interpreted as cell fractions, they showed high correlation with the true cell proportions.
We have designed an interactive web-based portal using Rshiny technology for a comprehensive validation of two most popular immune cell deconvolution methods xCell and CIBERSORT on publicly available data. The app could be found here: https://emdserono1.shinyapps.io/Immune_Cell_Deconvolution_Validation/
- https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE93777
- Newman, A. M. et al. Robust enumeration of cell subsets from tissue expression profiles. Nat Methods 12, 453-457, doi:10.1038/nmeth.3337 (2015).
- Aran, D., Hu, Z. & Butte, A. J. xCell: digitally portraying the tissue cellular heterogeneity landscape. Genome Biol 18, 220, doi:10.1186/s13059-017-1349-1 (2017).
