Curating Regulatory Networks from Literature

We curated multiple experimentally derived networks of regulatory interactions from published databases and the literature (Supplementary File S3) to serve as gold standards for our network inference algorithms. These experiments are typically based on ChIP-chip, ChIP-seq or regulator perturbation followed by global transcriptome profiling. We obtained multiple networks based on ChIP and TF perturbation experiments for each organism and cell type. When multiple ChIP or perturbation interactions were available we took a union of the networks. We refer to the ChIP derived gold standard as “ChIP” and the perturbation derived gold standard as “Perturb”. Finally, we took the intersection of these two unions, as the third primary gold standard networks for our evaluations of network accuracy (ChIP+Perturb).
Finally, we created a fourth ESC specific gold standard network from the primary literature (Zhou et al. 2007 (link); Kim et al. 2008 (link); Young 2011 (link); Buganim et al. 2012 (link); Dunn et al. 2014 (link); Xu et al. 2014 (link); Malleshaiah et al. 2016 (link)) by conducting a literature survey of gene regulatory networks for the ESC state. Regulatory edges from the publications were manually extracted from network figures and further curated by a stem cell biologist (see Acknowledgments). The specific publications and figures used to create our curated gold standard are in Supplementary Table S5.