Background Transcription factor(TF) interactions are known to regulate gene expression in eukaryotes via TF regulatory modules(TRMs). Such interactions can be formed due to co-localizing TFs binding proximally to each other in the DNA sequence or between distally binding TFs via long distance chromatin looping. While the former type of interaction has been characterized extensively, long distance TF interactions are still largely understudied. Furthermore, most prior approaches have focused on characterizing physical TF interactions without accounting for their effects on gene expression regulation. Understanding how TRMs influence gene expression regulation could aid in identifying diseases caused by disruptions to these mechanisms. In this paper, we present a novel neural network based approach to detect TRM in the GM12878 immortalized lymphoblastoid cell line.
Results We estimated main effects of 149 individual TFs and interaction effects of 48 distinct combinations of TFs for their influence on gene expression based on the neural networks trained to predict gene expression using multi-omics TF regulatory features. We identified several well-known TF interactions and discovered multiple previously uncharacterized TF interactions within our detected set of TRMs. We further characterized the pairwise TF interactions using long distance chromatin looping and motif co-occurrence data. We found that nearly all the TFs constituting TRMs detected by our approach interacted via chromatin looping, and that these TFs further interact with promoters to influence gene expression through one of four possible regulatory configurations.
Conclusion We have detected TRMs using neural network models based on regulatory features. We have also described these TRMs based on their regulatory potential along with presenting evidence for the possibility of TF interactions forming the TRMs occurring via chromatin looping.