Fig. 1

Workflow for building prediction models using multi-omics GRNs. ChIP-seq data for 153 TFs (GM12878) and 382 TFs (K562) having peaks passing the optimal irreproducible discovery rate (IDR) threshold defined by ENCODE were mapped to the regulatory region of each gene to define TFBS. The most distant CTCF peaks within a 50 Kb window upstream and downstream of the gene body were used to demarcate regulatory boundaries. Statistically significant TFBS from these regions were identified by FIMO and TEPIC based TF-TG affinity scores were calculated. PANDA GRNs were then generated using weighted and unweighted adjacency matrices. PPI data from BioGRID corresponding to TFs for each cell lines and cell line specific co-expression were obtained from GEUVADIS (GM12878) and ENCODE (K562). Elastic Net (ENET)-based regularized regression models were built from the resulting input features to predict log FPKM values (gene expression) of independent datasets for the two cell lines