We develop a novel ChIP-seq based Gibbs Sampling approach for cis-regulatory Module inference (ChIP-GSM). ChIP-GSM iteratively samples read tags of multiple TFs and their possible combinations under a Gibbs sampling framework. If read counts are from bindings at gene promoter regions, ChIP-GSM will identify promoter-focused TF modules; or if from enhancer regions, ChIP-GSM will provide a list of modules functional at enhancers. In our study, we do find difference between TF combinations at promoter and enhancer regions. For a general use, if there is no region preference, all binding signals from the whole genome can be jointly studied by ChIP-GSM.

The regulatory effect of TF modules (instead of individual TFs) was directly modeled by ChIP-GSM. Specifically, we use a weight-based read tag tossing approach to identify binding events for multiple TFs simultaneously, where any binding event with a relatively low read count can be well captured by assigning a weight much higher than that of a background region. Such ‘weak’ bindings will help identify TF associations across multiple regions and further highlight their cooperative action in a cell type-specific manner.