Fig. 1

Inference and analysis of GRNs using netZoo. YARN normalizes gene expression (RNA-Seq) data to account for differences between tissues. Then, a first group of methods uses normalized gene expression data to infer gene regulatory networks (PANDA, PUMA, OTTER, LIONESS, SPIDER, EGRET) to reconstruct GRNs using multiple genomic modalities. The input data used for PANDA and OTTER are normalized RNA-Seq data to build gene coexpression networks, PPI network such as STRINGdb to build TF interaction networks, and a prior knowledge TF motif network built on scanning TF position weight matrices in promoter region of target genes. We refer to these three input networks as the core input data that may be shared by groups of methods. In addition to this core input, SPIDER uses DNase-Seq chromatin accessibility data to constrain predictions to open regions of the genome. Instead of using TF motif network, PUMA employs miRNA target gene prediction data from tools such as TargetScan and miRanda as a prior knowledge network to seed inference of miRNA regulation networks. EGRET uses data from DNA sequence to first identify variants in TF binding sites and compute their effect on target gene regulation by combining these mutation data with the core input data. DRAGON builds multi-omic, partial correlation-based networks that can use data such as RNA-seq, methylation status, protein levels, and chromatin accessibility. A second group (CONDOR, ALPACA, CRANE) identifies communities in the networks (CONDOR), finds differential community structures between two networks of interest (ALPACA), and estimates the significance of differences between modules (CRANE). Finally, MONSTER estimates a transition matrix between two networks representing an initial and a final state, and SAMBAR de-sparsifies mutation data using biological pathways. Overlapping methods share the same input data. SNP, single nucleotide polymorphism; PPI, protein-protein Interaction network; miRNA, microRNA