Modeling transcription factor combinatorics in promoters and enhancers

We propose a new approach (TFcoop) that takes into account cooperation between transcription factors (TFs) for predicting TF binding sites. For a given a TF, TFcoop bases its prediction upon the binding affinity of the target TF as well as any other TF identified as cooperating with this TF. The set of cooperating TFs and the model parameters are learned from ChIP-seq data of the target TF. We used TFcoop to investigate the TF combinations involved in the binding of 106 different TFs on 41 different cell types and in four different regulatory regions: promoters of mRNAs, lncRNAs and pri-miRNAs, and enhancers. Our experiments show that the approach is accurate and outperforms simple PWM methods. Moreover, analysis of the learned models sheds light on important properties of TF combinations. First, for a given TF and region, we show that TF combinations governing the binding of the target TF are similar for the different cell-types. Second, for a given TF, we observe that TF combinations are different between promoters and enhancers, but similar for promoters of distinct gene classes (mRNAs, lncRNAs and miRNAs). Analysis of the TFs cooperating with the different targets show over-representation of pioneer TFs and a clear preference for TFs with binding motif composition similar to that of the target. Lastly, our models accurately distinguish promoters into classes associated with specific biological processes.