GraphLooper: Predicting Chromatin Loops Based on Hierarchical Multi-View Graph Pooling Method

Chromatin loops serve as fundamental functional units of three-dimensional genome organization, playing a pivotal role in regulating gene expression and maintaining genomic spatial organization. Accurate identification of these fine-scale structures is crucial for advancing our understanding of cellular biological processes and the mechanisms underlying disease. However, due to the inherent complexity and dynamic of chromatin interactions, existing prediction methods often fail to adequately characterize and comprehensively capture their multi-dimensional features. To address this limitation, we present GraphLooper, a novel framework based on hierarchical multi-view graph pooling for training and inference on large-scale data. GraphLooper first transforms Hi-C data into a graph-structured representation and integrates multi-dimensional epigenomic features, constructing a comprehensive system for chromatin interaction representation. By introducing a hierarchical pooling mechanism, GraphLooper effectively aggregates multi-scale features, significantly enhancing the model's representation learning capabilities. Systematic evaluation across various cell lines demonstrates that GraphLooper not only surpasses existing state-of-the-art methods in prediction accuracy but also exhibits strong generalization performance. Notably, it demonstrates exceptional ability in capturing long-range chromatin interactions, which are critical for remote gene regulation through precise spatial organization.