quality surrogate variables (qSV) (10.1073/pnas.1617384114) revealed qSVs may already adjust for variation in cell type composition. Correlation analysis of the cell type proportions vs. However, the effect size of the differences is minimal. Between samples from male and female donors 6 out of 20 t-tests are significant (testing for ten cell types, in two brain regions p.bonf < 0.05). The Bisque estimated cell type proportions showed a difference in means across diagnoses (MDD, Bipolar, control) for either brain region (AMY, sACC) in 22 out of 60 pairwise t-tests (testing ten cell types, in two brain regions, between three diagnoses p.bonf < 0.05). Bisque also returns cell fraction estimates that were concordant with expected regional variability (10.1101/2021.00). We observed that MuSiC is highly sensitive to the set of marker genes, while Bisque is robust. The resulting set of marker genes overlaps eight known marker genes (10.1038/s4158-2) and identifies new data driven marker genes (96%). The set of marker genes we identified is highly specific, confirmed by visualizing violin plots of gene expression over cell type, heat maps of pseudo-bulked cell type from each donor, and t-tests comparing against all other cell types (FDR < 5%). We will focus here on the deconvolution of the MDDseq data set containing 1,091 samples, with 704 (459 MDD, 245 Bipolar) cases and 387 control samples, from the sACC and AMY (Synapse syn22276064). We performed deconvolution on eight, in-house, post-mortem, human brain, bulk RNAseq datasets spanning 5,787 samples, in seven brain regions (DLPFC, sACC, AMY, HPC, Caudate, Dentate Gyrus, and Nucleus Acumbens), and four diagnoses (MDD, Bipolar, SCZD, and control), as well as the GTEx v8 brain dataset. Deconvolution was performed with MuSiC version 0.2.0 and the ReferenceBasedDecomposition function from BisqueRNA version 1.0.4, using the use.overlap = FALSE option. MuSiC claims to be good for differentiating between closely related cell types, and Bisque showed the highest performance in their benchmark analysis on DLPFC data. The top 25 genes for each cell type were selected. To achieve this, we developed a marker selection strategy: calculating the ratio of the mean expression of each gene in the target cell type over the highest mean expression of that gene in a non-target cell type. To give the algorithms the best change of accuracy, we selected for highly specific genes to each cell type, ideally only expressed in the target cell type. The ten cell types considered in the deconvolution of the tissue were Astrocytes (Astro), Endothelial (Endo), Macrophage (Macro), Microglia (Micro), Mural cells, Oligodendrocytes (Oligo), Oligodendrocyte Progenitor Cells (OPC), T cells, Excitatory Neurons (Excit), and Inhibitory Neurons (Inhib). (10.1101/2020.39) which includes 70,615 high-quality nuclei across five human brain regions: nucleus accumbens (NAc), amygdala (AMY), subgenual anterior cingulate cortex (sACC), hippocampus (HPC), and dorsolateral prefrontal cortex (DLPFC). Deconvolution seeks to estimate the composition of cell types in the sequenced tissue to examine biological differences between samples, or minimize confounding technical variables, such as differences in dissection, in downstream analysis.įor the reference single nucleus RNA-seq dataset we utilized the data set from Tran et al. Bulk RNA-seq from tissue homogenates obscures this complexity. Lieber Institute for Brain Development, Baltimore, Maryland, United Statesīackground: The brain contains a wide variety of structures, that in turn are made up of many different cell types with individual complex functions.
0 Comments
Leave a Reply. |