Characterizing collaborative transcription regulation with a graph-based deep learning approach

Human epigenome and transcription activities have been characterized by a number of sequence-based deep learning approaches which only utilize the DNA sequences. However, transcription factors interact with each other, and their collaborative regulatory activities go beyond the linear DNA sequence. Therefore leveraging the informative 3D chromatin organization to investigate the collaborations among transcription factors is critical. We developed ECHO, a graph-based neural network, to predict chromatin features and characterize the collaboration among them by incorporating 3D chromatin organization from 200-bp high-resolution Micro-C contact maps. ECHO predicted 2,583 chromatin features with significantly higher average AUROC and AUPR than the best sequence-based model. We observed that chromatin contacts of different distances affected different types of chromatin features’ prediction in diverse ways, suggesting complex and divergent collaborative regulatory mechanisms. Moreover, ECHO was interpretable via gradient-based attribution methods. The attributions on chromatin contacts identify important contacts relevant to chromatin features. The attributions on DNA sequences identify TF binding motifs and TF collaborative binding. Furthermore, combining the attributions on contacts and sequences reveals important sequence patterns in the neighborhood which are relevant to a target sequence’s chromatin feature prediction.Author summary: Human transcription activities are regulated by chromatin features including transcription factor binding, histone modification, and DNase I hypersensitive site. Recently many computational models are proposed to predict chromatin features from DNA sequence. However, human genome has a complex and dynamic spatial organization, and chromatin loops form and bring regulatory elements that lie far apart on the genomic sequence into spatial proximity so that transcription factors which bind far apart may interact with each other. Therefore, to investigate the collaborations among chromatin features, utilizing 3D chromatin organization is critical. In this work, we propose a graph neural network model to predict chromatin features in the light of 200bp-resolution Micro-C contact maps which capture fine-scale chromatin contacts well. Furthermore, by interpreting the model, we identify important chromatin contacts which contribute to chromatin feature prediction, and characterize the collaborations among these chromatin features, which helps researchers understand transcription factor collaborative binding mechanisms.