Fig 1.
Relationship between correlation bias and library-size variation.
a) Library-size variance-dependent correlation inflation in null data. Scatter plot demonstrates systematic overestimation of co-expression (Pearson correlation coefficients between gene pairs) (y-axis) with increasing mean expression levels (x-axis) under varying library size variance (Variance size factors). Variance size factors scale the variance of library size of the simulated data. The higher the variance size factor, the greater the library size variation in the simulated data. b) Library-size variance-dependent correlation absolute bias. Scatter plot showing the scale factor (variance of library size/mean of library size) and average absolute bias of the estimated correlation. The higher the variance size factor, the greater the library size variation in the simulated data.
Table 1.
PPI enrichment among the top 10,000 correlated gene pairs across cell types. Values denote the number of gene pairs exhibiting significant PPI enrichment for each method within each cell type. Paired Wilcoxon test comparing LSMetacell with all benchmark methods across cell types yielded P = 0.031 for every comparison.
Fig 2.
Preservation of co-expression modules constructed by different methods across major cell types.
Co-expression modules were constructed separately for excitatory neurons (EN), inhibitory neurons (IN), oligodendrocytes (OC), astrocytes (AC), and microglia (MC). The preservation of these modules was subsequently evaluated in a fully independent single-cell sequencing dataset using the Z-Summary statistic. A Z-Summary > 10 indicates strong evidence of preservation, while a Z-Summary > 2 suggests moderate preservation.
Fig 3.
Microglia gene co-expression modules.
Module names are prefixed with “MC” (abbreviation for microglia) followed by the color label. a) UMAP representation of the co-expression network visualizes individual genes (kMEs > 0.2) with eigengene of each module indicated by larger circles colored by the module color. Positive module eigengene correlations > 0.5 are indicated by grey lines connecting the eigengenes. b) Modules-phenotype association analysis. Correlation between the module eigengene and phenotypes. Numbers indicate correlation coefficients and p-values. Braaksc: Braak Stage is the semiquantitative measure (from no effect on neocortex to severity effect) of severity of neurofibrillary tangle (NFT) pathology. Ceradsc: CERAD score is the semiquantitative measure of neuritic plague (from severity to no effect). Cogdx: Clinical consensus diagnosis of cognitive status at time of death (from no cognitive impairment to dementia). Dcfdx_lv: Clinical diagnosis of cognitive status at last visit (from no cognitive impairment to dementia). c) The biological process gene ontology enrichment analysis of genes. GeneRatio is the ratio between genes of interest in the gene set and total genes of interest. Dot color represents the adjusted p-value (Benjamini–Hochberg method) of enrichment analysis.