NLM DIR Seminar Schedule

Abstract:

Gene regulation in eukaryotes mainly involves transcription factors (TFs). These proteins bind to regulatory DNA elements such as enhancers and determine the amount and timing of target gene expression. Mutations at TF binding sites (TFBSs) are associated with complex human diseases and traits. Consequently, accurately identifying TFBSs is crucial to pinpoint causal variants.
Computational state-of-the-art algorithms typically use position weight matrices (PWMs) to identify TFBSs in the human genome; however, these algorithms produce too many false positives. Here, we use TREDNet—a deep learning model developed in our research group—to identify TFBSs in HepG2 cell line enhancers accurately. We identify TFBSs at enhancer regions that would damage the enhancer upon mutation, called positive active regions (PARs), and that would strengthen the enhancer upon mutation, called negative active regions (NARs). We found that the NARs are more GC enriched than the PARs. Clustering analysis of the TFBSs at PARs revealed ~10 groups of binding sites. In addition, analysis of TF pair co-occurrence revealed that the forkhead box (FOX) family of TFs is prevalent in PAR regions.