Benchmark of cellular deconvolution methods using a multi-assay dataset from postmortem human prefrontal cortex
Louise A. Huuki-Myers, Kelsey D. Montgomery, Sang Ho Kwon, Sophia Cinquemani, Nicholas J. Eagles, Daianna González‐Padilla, Sean K. Maden, Joel E. Kleinman, Thomas M. Hyde, Stephanie C. Hicks, Kristen R. Maynard, Leonardo Collado‐Torres
Abstract
Cellular deconvolution of bulk RNA-sequencing data using single cell/nuclei RNA-seq reference data is an important strategy for estimating cell type composition in heterogeneous tissues, such as the human brain. Here, we generate a multi-assay dataset in postmortem human dorsolateral prefrontal cortex from 22 tissue blocks, including bulk RNA-seq, reference snRNA-seq, and orthogonal measurement of cell type proportions with RNAScope/ImmunoFluorescence. We use this dataset to evaluate six deconvolution algorithms. Bisque and hspe were the most accurate methods. The dataset, as well as the Mean Ratio gene marker finding method, is made available in the DeconvoBuddies R/Bioconductor package.