Fig 1.
Overview of the CONDOR algorithm.
All possible SNP-gene pairs from an appropriate data set are considered in an eQTL analysis. Both cis- and trans-acting eQTLs (FDR < 0.1) are used to construct a bipartite network linking SNPs and genes. The resulting network structure is then analyzed, first globally to understand its overall structure and to identify network “hubs.” Then the community structure of the bipartite network is determined, each community is subject to functional enrichment analysis, and a core score is calculated to identify those SNPs most likely to disrupt individual communities.
Fig 2.
Quantile-quantile plot for 13,333,199 cis- and 17,228,062,483 trans-eQTL p-values.
Fig 3.
SNPs and genes display broad-tailed degree distributions.
The degree distribution, with the frequency of node degree plotted on a log-log scale, is shown for SNPs (a) and genes (b) in all connected components with more than 5 SNPs and 5 genes in the bipartite eQTL network.
Fig 4.
Degree distributions for NHGRI-GWAS (red) and all (black) SNPs.
NHGRI-GWAS SNPs tend not to be global network “hubs,” which are located in the far-right tail of the distribution. The highest degree NHGRI-GWAS SNP was connected to 10 genes.
Fig 5.
eQTLs show strong community structure.
(a) Plot of the communities within the bipartite eQTL network. The nodes (genes and SNPs) in each community form a ring, with the link density within each ring visibly darker than links between communities. (b) Links within communities (colored points) are shown along the diagonal, with links that go between communities in black. Community IDs are plotted along the x-axis.
Fig 6.
Communities comprise SNPs and genes from multiple chromosomes.
Number of different chromosomes in each community based on (a) SNP and (b) gene locations.
Fig 7.
NHGRI-GWAS SNPs have higher core scores than non-GWAS SNPs based on Kolmogorov-Smirnov test statistics.
Histogram of Kolmogorov-Smirnov test statistics comparing the distribution of Qih scores for sets of randomly relabeled NHGRI-GWAS/non-GWAS SNPs. The KS test statistic for the true labeling is in red. The permutation p-value associated with the KS test is P < 10−5 given 105 permutations.
Fig 8.
NHGRI-GWAS SNPs have higher core scores than non-GWAS SNPs based on Wilcoxon test statistics.
Histogram of Wilcoxon test statistics comparing the distribution of Qih scores for sets of randomly relabeled NHGRI-GWAS/non-GWAS SNPs. The Wilcoxon test statistic for the true labeling is in red. The permutation p-value associated with the Wilcoxon test is P < 10−5 given 105 permutations.
Fig 9.
The majority of COPD Network GWAS SNPs are annotated for functional impact.
Of the 30 SNPs that are eQTLs in the LGRC network and also associated with COPD (FDR < 0.05), 15 are likely to affect transcription factor (TF) binding and linked to the expression of a target gene (a score of 1b, d, or f), 2 have evidence of TF binding or a DNase peak (a score of 5), and 11 are located in a motif hit (a score of 6) according to RegulomeDB [37].
Fig 10.
The median core score for COPD Network GWAS SNPs is higher than for non-significant SNPs.
The median core score for the 30 FDR-significant COPD GWAS SNPs (FDR < 0.05, left) is 20.3 times higher than the median core score for the non-significant SNPs (FDR ≥ 0.05, right).