2.6. Integrative networks of genomic and transcriptomic data

To study how genomic and liver transcriptomic variation contributes to hepatic GCase activity variability in the 27 HMDP strains, we employed Mergeomics v1.18 [24]. To build Bayesian networks of integrative omics underlying GCase activity, two modules of Mergeomics are required: a) marker set enrichment analysis (MSEA) and b) weighted key driver analysis (wKDA). MSEA requires the following data inputs: 1) EMMA GWAS results: i) marker-GCase activity association (marker-value) and ii) gene-marker mapping file (gene-marker); 2) functionally related gene sets (module-gene), which are preloaded in Mergeomics. These results are integrated through the package algorithm to find sets of genes associated with GCase activity. The parameter settings of the MSEA module included: i) type of permutation at the gene level. ii) minimum (10) and maximum (500) number of genes in the sets. iii) the minimum and maximum overlap ratio between sets of genes associated with disease/trait = 0.33 (33% overlap). iv) the number of gene or marker permutations = 2000 and finally v) the MSEA FDR cutoff was ≤25% [25], this analysis calculates the Benjamini-Hochberg FDR [26].

To identify key driver (KD) genes, which are defined as the gene hubs most significantly associated with other genes in the network, we used wKDA [24]. The wKDA module takes input data from the MSEA results generated in the previous step and a defined liver tissue Bayesian network corresponding to human and rodent expression datasets of earlier studies [27]. The parameters for running wKDA included i) Search depth of wKDA = 1, which means that we search for key-drivers whose immediate neighborhood is enriched for MSEA significant genes, ii) the edge type of wKDA = incoming and outgoing directionality, iii) the minimum overlap, is the threshold above which hubs will be designated as co-hubs, of wKDA = 0.33, and iv) the edge factor of wKDA = 0.5, which means an unweight network. This module projected sets of genes associated with liver GCase activity onto a Bayesian liver network, representing seemingly causal relationships between genes and KD genes [27]. We ran both Mergeomics modules in the R package [19].