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

To generate a dataset that can be used as a reference for postmortem human brain bulk RNA-sequencing deconvolution, several orthogonal assays were completed on individual fresh frozen tissue blocks of human dorsolateral prefrontal cortex (DLPFC). These assays included spatially-resolved transcriptomics (Visium Spatial Gene Expression, 10x Genomics), single nucleus RNA-sequencing (snRNA-seq, Single Cell 3' Gene Expression, 10x Genomics), single molecule fluorescent in situ hybridization data (smFISH) combined with immunofluorescence (IF) using RNAScope/IF (Advanced Cell Diagnostics) , and bulk RNA-sequencing (RNA-seq) across six different RNA library types. Such datasets are needed to resolve "challenges and opportunities to computationally deconvolve heterogeneous tissue with varying cell sizes using single-cell RNA-sequencing datasets" (doi: 10.1186/s13059-023-03123-4).DLPFC data was generated from ten donors across three positions: anterior, middle, and posterior. Spatially-resolved transcriptomics data for all 30 samples and single nucleus RNA-sequencing (snRNA-seq) for 19 samples was generated as previously described (doi: 10.1101/2023.02.15.528722), using the Visium and Chromium assays by 10x Genomics, respectively.RNAScope/IF imaging experiments used cell type markers to identify and quantify the proportions of 6 broad cell types (excitatory neurons, inhibitory neurons, astrocytes, microglia, oligodendrocytes, endothelial cells). These cell type proportions, when generated across serial sections, can be used to benchmark RNA-seq deconvolution algorithms. RNAScope/IF also provides nucleus size data as the nucleus is stained with DAPI. In this study, RNAScope/IF images were generated from a subset of 30 tissue blocks (doi: 10.1101/2024.02.09.579665)."Data-driven identification of total RNA expression genes for estimation of RNA abundance in heterogeneous cell types highlighted in brain tissue" identifies genes (doi: 10.1186/s13059-023-03066-w) that can be used to estimate total RNA content for different cell types. In addition to cell type markers described above, AKT3, a total RNA expression gene (TREG), was included in the RNAScope/IF experiments.Postmortem human brain bulk RNA-seq data can be generated across different RNA fractions (cytosolic vs. nuclear vs. total), and either with polyA or RiboZeroGold RNA-seq library preparation types, which target different pools of RNA molecules. In this study, bulk RNA-seq with either polyA or RiboZero for three different cell fractions (cytosolic, nuclear, or total) were generated across 19 of the tissue blocks (19 * 2 * 3 = 114; n=113 total bulk RNA-seq samples; data could not be generated for 1 polyA nuclear fraction sample).This NCBI SRA entry is for the bulk RNA-seq FASTQ files. These data were generated to benchmark and develop bulk RNA-seq deconvolution algorithms that use snRNA-seq reference data. See "benchmark of cellular deconvolution methods using a multi-assay reference dataset from postmortem human prefrontal cortex" for more details (doi: 10.1101/2024.02.09.579665).