SDP-ChroNet: A shared dilated pooling network for chromatin interaction prediction

A deep understanding of chromatin interactions is crucial for elucidating the mechanisms of gene expression regulation. Although various computational methods have been proposed to predict chromatin interactions and address the limitations and high cost of high-throughput experimental techniques, their application prospects remain limited. A well-recognized challenge in this field lies in the severe class imbalance inherent in chromatin interaction data, where the disparity between positive and negative samples is known to bias classifiers toward the majority class, thereby contributing to the generally low predictive performance of existing models and hindering their widespread application in biomedical research.models and hinders their widespread application in biomedical research. To overcome this challenge, we propose a novel deep learning framework, termed SDP-ChroNet, which is designed to efficiently capture local motifs and long-range dependencies in long sequences via a Shared Residual Dilated Unit (SRDU), combined with genomic features, to identify chromatin interactions. We evaluated SDP-ChroNet on multiple datasets and benchmarked its performance against current state-of-the-art methods. Experimental results demonstrate that SDP-ChroNet consistently outperforms existing methods in the task of chromatin interaction prediction. Moreover, we propose a novel computational method to quantify the conditionally shared effects of genomic features across different cellular contexts, thereby providing computational evidence for potential regulatory signals underlying cross-domain shifts in chromatin interaction prediction. We anticipate that SDP-ChroNet will serve as a robust tool for future studies that seek to identify chromatin interactions.