CLAMP: predicting specific protein-mediated chromatin loops in diverse species with a chromatin accessibility...

Emerging DNA language models provide powerful tools to address the challenge of accurately predicting chromatin loops, fundamental structures governing 3D genome organization and gene regulation. Here we present CLAMP, which utilizes a deep language model pre-trained on broad cross-species chromatin accessibility data. CLAMP achieves superior performance compared to existing methods in predicting specific protein-mediated loops across 10 species, 18 proteins, and 24 cell types. CLAMP incorporates a novel CoVE explainer that reveals context-dependent genomic feature contributions, providing insights into the features driving predictions. CLAMP predictions effectively identify functionally significant chromatin loops and associated biological pathways.