Benchmark of cellular deconvolution methods using a multi-assay reference dataset from postmortem human prefrontal cortex.

BACKGROUND: Cellular deconvolution of bulk RNA-sequencing (RNA-seq) data using single cell or nuclei RNA-seq (sc/snRNA-seq) reference data is an important strategy for estimating cell type composition in heterogeneous tissues, such as human brain. Several deconvolution methods have been developed and they have been previously benchmarked against simulated data, pseudobulked sc/snRNA-seq data, or cell type proportions derived from immunohistochemistry reference data. A major limitation preventing the improvement of deconvolution algorithms has been the lack of highly integrated datasets with orthogonal measurements of gene expression and estimates of cell type proportions on the same tissue block. The performance of existing deconvolution algorithms has not yet been explored across different RNA extraction methods (e.g. cytosolic, nuclear, or whole cell RNA), different library preparation types (e.g. mRNA enrichment vs. ribosomal RNA depletion), or with matched single cell reference datasets.

RESULTS: A rich multi-assay dataset was generated in postmortem human dorsolateral prefrontal cortex (DLPFC) from 22 tissue blocks. Assays included spatially-resolved transcriptomics, snRNA-seq, bulk RNA-seq across six RNA extraction and RNA-seq library combinations, and orthogonal cell type measurements via RNAScope/Immunofluorescence (RNAScope/IF). The Mean Ratio method was developed for selecting cell type marker genes for deconvolution and is implemented in the DeconvoBuddies R package. Five extensively benchmarked computational deconvolution algorithms were evaluated in DLPFC across six RNA-seq combinations and predicted cell type proportions were compared to those measured by RNAScope/IF.

CONCLUSIONS: We show that Bisque and hspe are the top performing methods with performance dependent on the RNA-seq library preparation conditions. We provide a multi-assay resource for the development and evaluation of deconvolution algorithms.