A prior network derived from ATAC-seq data improves TRN recovery of gold-standard TF–gene interactions. (A) Precision recall of Th17 TRNs. The left two panels enable comparison of TRNs built from ChIP versus Th17 ATAC priors, quantified by precision recall of the KO GS (25 TFs, 8875 interactions); insets display AUPR. The performance of several TRNs are plotted for each prior, based on Inferelator method. (LS) mLASSO-StARS; (BB) BBSR-BIC; (m) TF mRNA; (TFA) prior-based TFA; (+) indicates strength of prior reinforcement. Random and “no prior” TRNs serve as references. The right panel shows precision recall of the KO-ChIP GS (nine TFs, 2375 edges) for TRNs built from the Th17 ATAC prior. (B) Number of targets per TF in the GSs. Targets per TF are limited to the 3578 considered by the model. (C) TF-specific TRN performance. For each GS, AUPRs were calculated for each TF individually. TF-specific performance of TRNs is quantified as the log₂ fold-change between AUPR of the TRN model relative to random. +, m, and TFA are as in A.