Systematic evaluation of transcriptomics-based deconvolution methods and references using thousands of clinical samples

Abstract

Estimating cell type composition of blood and tissue samples is a biological challenge relevant in both laboratory studies and clinical care. In recent years, a number of computational tools have been developed to estimate cell type abundance using gene expression data. Although these tools use a variety of approaches, they all leverage expression profiles from purified cell types to evaluate the cell type composition within samples. In this study, we compare 12 cell type quantification tools and evaluate their performance while using each of 10 separate reference profiles. Specifically, we have run each tool on over 4000 samples with known cell type proportions, spanning both immune and stromal cell types. A total of 12 of these represent in vitro synthetic mixtures and 300 represent in silico synthetic mixtures prepared using single-cell data. A final 3728 clinical samples have been collected from the Framingham cohort, for which cell populations have been quantified using electrical impedance cell counting. When tools are applied to the Framingham dataset, the tool Estimating the Proportions of Immune and Cancer cells (EPIC) produces the highest correlation, whereas Gene Expression Deconvolution Interactive Tool (GEDIT) produces the lowest error. The best tool for other datasets is varied, but CIBERSORT and GEDIT most consistently produce accurate results. We find that optimal reference depends on the tool used, and report suggested references to be used with each tool. Most tools return results within minutes, but on large datasets runtimes for CIBERSORT can exceed hours or even days. We conclude that deconvolution methods are capable of returning high-quality results, but that proper reference selection is critical.

Evaluating performance and applications of sample-wise cell deconvolution methods on human brain transcriptomic data

“…The performance of methods for estimating cell proportions has been previously evaluated (12)(13)(14)(15)(16). Studies have evaluated the accuracy of estimated cell proportions with data from the brain and other tissues.…”

Transcriptomic profiling of peripheral blood cells in HPV‐associated carcinoma patients receiving combined valproic acid and avelumab

“…The CIBERSORTx from Newman Lab ( https://cibersortx.stanford.edu/ ) [ 43 ] is used to deconvolute immune cell fractions and abundances from the isolated PBMCs' bulk RNA‐Seq data [ 44 ]. The TPM (Transcripts per million) files, representative of the gene length‐normalized expression data, were uploaded as the mixture of gene files to the CIBERSORTx website and compared with software incorporated LM22 signature matrix file.…”

Molecular Group and Correlation Guided Structural Learning for Multi-Phenotype Prediction

“…We demonstrate the potential use and performance of Brilliant using the real data from our in-house multi-omics Epigenetic Variation and Childhood Asthma in Puerto Ricans (EVAPR) study [10] and the Framingham Heart Study (FHS) [24, 22]. In the EVAPR study, we collect phenotype, bulk RNA-sequencing (RNA-seq) gene expression and DNA methylation (DNAm) data from Puerto Rican children aged 9-20 years.…”