Fig 1.
Regional association plot of the chromosome 4q genomic region bounded by COPD risk loci near BTC and HHIP.
Regional association plot of the 4q genomic region bounded by COPD risk loci near BTC and HHIP from the Sakornsakolpat et al. 2019 COPD GWAS study [6]. P-values are two-sided based on Wald statistics (35,735 cases and 222,076 controls). The loci are labeled according to the gene with a transcription start site closest to the top GWAS variant at each locus. The red line corresponds to genome-wide significance (p = 5 x 10−8).
Fig 2.
The first box shows how the COPD and the Differential partial correlation networks have been built. Given a source of biological information, such as a PPI network or set of functional annotations, for every gene i and j (gi,gj) we extract the gene specific regularization vector λi,j, that contains the distance between gene gi and gj to all of the other genes, in our case the inverse of the personalized PageRank algorithm (more details in S1 Text). For every pair of genes (gi,gj), we compute the Gene-Specific Ridge Partial Correlation. Finally, to build the partial correlation network we run a t-test, using permutation and bootstrap procedures, respectively. The second box contains the analysis done on the two networks from (1) the COPD subjects and (2) a differential network of COPD patients vs Healthy controls. Both analyses provide insights into the presence of a potential co-regulatory network in this genomic region.
Table 1.
Values represent mean and (standard deviation) for continuous variables and count (percentage) for binary variables. The values with (*) are statistically different between cases and controls, p-value <10−3. 34 control patients and 28 case patients have missing data for smoking status.
Fig 3.
Absolute Partial Correlation Distribution.
Distributions of the absolute value of the partial correlations between genes in the 4q COPD risk region using the LTRC RNA-seq data, with different parameter settings, including min_lambda (magnitude of the regularization parameter) and n_genes (number of controlling genes).
Fig 4.
COPD Candidate Gene (CCG) partial correlation distributions.
Boxplot of CCG partial correlation distribution over the 29 different parameter combinations in (a) COPD case and (b) control subjects. NAP1L5 has been removed because it did not show any significant partial correlation. A gene pair is identified by the position on the x-axis and the color of the distribution.
Fig 5.
Partial Correlation Network topologies.
(a) shows the network density (fraction of all possible gene pairs that have an edge) and (b) shows the size of the largest connected component (the highest number of nodes that are connected to each other directly or through other nodes). Each network is identified by the set of parameters used to compute the Partial Correlation (n_genes, min_lambda). n_genes is the number of controlling genes and min_lambda is the normalization parameter, which are used to compute the partial correlation.
Fig 6.
Partial Correlation Co-occurrence clustering heatmap.
Each entry of the matrix represents how many times two genes appear in the same module across partial correlation networks with different parameter settings. Hierarchical clustering of genes based on their co-occurrence across the partial correlation networks identified 10 clusters. CCG are shown on the two axes.
Fig 7.
COPD Candidate Genes are present in the differential partial correlation network.
An edge between two genes is present if the two genes are significantly differentially partially correlated (FDR<0.1). (a) The number of significant edges obtained for each set of parameters. (b) The histogram showing how frequently an edge was significant in the partial correlation networks estimated using different parameters. The edge with the highest frequency is (CXCL10-CXCL11). (c) The network obtained by edges present in the partial correlation with >1 controlling genes (n_genes). The edge color is the average partial correlation difference obtained with all of the examined parameters. Blue edges represent pairs of genes that are more positively correlated in the control subjects, red edges represent pairs of genes that are more correlated in COPD subjects. COPD Candidate Genes (CCG) are highlighted with a dark green outline.
Table 2.
Genes in the differential partial correlation network are involved in cell-based knock-out experiments.
The table shows the enrichment analysis of the nodes of the Differential Partial Correlation Network, COPD vs healthy controls, with the LINCS L1000 project perturbation gene sets. The first column shows the name of the knockout experiment, then the p-values and the adjusted p-values for multiple hypothesis testing. Finally, the gene column highlights the genes there are present in the differential network and in the knock-out gene set.