Fig 1.
Schematic of a multilayer gene co-expression network.
The intralayer edges, shown by the solid lines, represent co-expression. The interlayer edges, shown by the dashed lines, connect the same gene across layers.
Table 1.
Similarity between the highly variable genes and the highly expressed genes in each tissue.
Fig 2.
Determination of the resolution parameter value by CHAMP.
(a) Multilayer correlation matrix. (b) Multilayer network obtained by graphical lasso. The convex hull of the lines in the (γ, Q) plane, each of which corresponds to a node partitioning, is a piecewise linear curve with the transition values indicated by a cross and change in the line color. Each line segment corresponds to the optimal node partitioning in the corresponding range of γ.
Fig 3.
Composition of each community by layer, i.e., tissue.
(a) Multilayer correlation matrix, γ = 1. (b) Multilayer correlation matrix, γ = 3. (c) Unweighted multilayer network obtained by graphical lasso, γ = 1. (d) Unweighted multilayer network obtained by graphical lasso, γ = 3. (e) Weighted multilayer network obtained by graphical lasso, γ = 1. (f) Weighted multilayer network obtained by graphical lasso, γ = 3. The darker shades indicate nodes corresponding to genes that only appear in one layer in the given community. The lighter shades indicate nodes corresponding to genes that appear in multiple layers in the community.
Table 2.
Z scores for the total intralayer weight within each community detected in the multilayer correlation matrix.
Table 3.
Specialist fraction and the corresponding tissue for each community detected in the multilayer correlation matrix and for each community detected in the multilayer networks obtained by graphical lasso.
Table 4.
Z score for the average distance between pairs of genes on each chromosome for the N = 203 genes.
Table 5.
Analysis of localization of genes in each community detected in the multilayer correlation matrix.
Fig 4.
Location of genes on chromosomes, colored by community.
There is a colored circle for each associated community pointing to each gene. Note that a gene can belong to more than one community, denoted by multiple colored circles next to each other horizontally pointing to the same gene. This figure allows us to visually see clusters of genes on specific chromosomes and their associated community.
Fig 5.
Expression and co-expression analysis of a cluster of genes in community 5 on chromosome 17.
The co-expression matrices for these genes in (A) skin and in (B) pancreas are shown. The average expression for each gene in these tissues is shown in the bar graphs. The location of these genes on chromosome 17 is shown in (C), with arrows (colored according to the associated tissue) pointing from putative regulatory elements to highly co-expressed genes. (D) The panel shows different measures of the regulatory potential of this genome section. From top to bottom: 1. H3K27AC modification to histone H3 within the region, which often correlates with activation of transcription and is associated with active enhancers in a given tissue available through ENCODE database [119]. 2. DNAse1 hypersensitivity sites. They are sections of the genome that are cut by DNAse1 enzyme. Given that the chromatin has to be “open” for the DNAse to access the sequence, the sequences that are cut by DNAse indicate open chromatin, which is in turn associated with regulatory activity. Data are available through ENCODE database [119]. 3. Enhancer/promoters. These are sequences that are predicted as enhancers (gray) and promoters (red) from the GeneHancer database [120]. 4. Established interactions between regulatory regions and genes as documented by GeneHancer database [120]. These data sets combined with our co-expression analysis provide a novel outlook into potential topologically associated domains that may be regulated by specific sequences in a tissue-specific manner.
Fig 6.
A schematic of SNPs in an enhancer region (gray box) that affect the expression of CELA3A (blue box) and CELA3B (orange box) in the pancreas and are associated with blood phosphate concentration.
(A) Expression levels of CELA3A and CELA3B, and blood phosphate concentration when the derived alleles for the putatively causal SNPs are absent. (B) The presence of the derived alleles for the putatively causal SNPs decreases the expression level of CELA3A, increases the expression level of CELA3B, and decreases the blood phosphate concentration.