Journal: Cell Reports Methods

Article Title: hdWGCNA identifies co-expression networks in high-dimensional transcriptomics data

doi: 10.1016/j.crmeth.2023.100498

Figure Lengend Snippet: Overview of the hdWGCNA workflow and application in the human prefrontal cortex (A) Schematic overview of the standard hdWGCNA workflow on a scRNA-seq dataset. UMAP plot shows 36,671 cells from 11 cognitively normal donors in the Zhou et al. human prefrontal cortex (PFC) dataset. ASC, astrocytes; EX, excitatory neurons; INH, inhibitory neurons; MG, microglia; ODC, oligodendrocytes; OPC, oligodendrocyte progenitor cells. (B) Density plot showing the distribution of pairwise Pearson correlations between genes from the single-cell (sc) expression matrix and metacell expression matrices with varying values of the K -nearest neighbors parameter K . (C) Expression matrix density (1, sparsity) for the sc, pseudo-bulk (pb), and metacell matrices with varying values of K in each cell type. (D) Heatmap of scaled gene expression for the top five hub genes by kME in INH-M6, EX-M2, ODC-M3, OPC-M2, ASC-M18, and MG-M14. (E) snRNA-seq UMAP colored by module eigengene (ME) for selected modules as in (D). (F) UMAP plot of the ODC co-expression network. Each node represents a single gene, and edges represent co-expression links between genes and module hub genes. Point size is scaled by kME. Nodes are colored by co-expression module assignment. The top two hub genes per module are labeled. Network edges were downsampled for visual clarity. (G) snRNA-seq UMAP as in (A) colored by MEs for the 10 ODC co-expression modules as in (F). (H) Module preservation analysis of the ODC modules in the Morabito et al. human PFC dataset. The module’s size versus the preservation statistic ( Z preservation) is shown for each module. Z < 5 , not preserved; 10 > Z ≥ 5 , moderately preserved; Z ≥ 10 , highly preserved.

Article Snippet: Metacell transcriptomic profiles were constructed separately for each of the 54 samples and each cell type using the hdWGCNA function MetacellsByGroups, aggregating 50 cells into one metacell.

Techniques: Expressing, Gene Expression, Labeling, Preserving

Journal: Cell Reports Methods

Article Title: hdWGCNA identifies co-expression networks in high-dimensional transcriptomics data

doi: 10.1016/j.crmeth.2023.100498

Figure Lengend Snippet: Runtime, memory usage, and performance of hdWGCNA (A and B) We ran the main co-expression network analysis functions of the hdWGCNA R package on 65,415 neuronal cells in a human brain dataset from 54 samples, and tracked the runtime (A) and memory usage upper bound (B) for different-sized subsets of the data ranging from 1,000 through 50,000 cells. (C) Violin plots showing distributions of EGAD neighbor-voting area under the receiver operating characteristic curve (AUC) scores in each of the cell-type-specific co-expression networks from the human PFC dataset. (D) Violin plots showing distributions of multifunctionality AUC scores in each of the cell-type-specific co-expression networks from the human PFC dataset. ASC, astrocytes; EX, excitatory neurons; INH, inhibitory neurons; MG, microglia; ODC, oligodendrocytes; OPC, oligodendrocyte progenitor cells. (E) Performance of the XGBoost regularized regression models used to predict gene expression based on the expression of the top 10 module hub genes for all 96 co-expression modules from the Zhou et al. human PFC dataset. Violin plots showing the test set root-mean-square error (RMSE) comparing the predicted expression with observed for each gene, split by each co-expression module. Modules are ordered within each cell type from lowest mean RMSE to highest.

Article Snippet: Metacell transcriptomic profiles were constructed separately for each of the 54 samples and each cell type using the hdWGCNA function MetacellsByGroups, aggregating 50 cells into one metacell.

Techniques: Expressing, Gene Expression