Accurate and efficient HiChIP interaction detection by modeling restriction enzyme cut site density as biological signal.

Abstract

HiChIP enables high-resolution chromatin interaction mapping, but existing methods generally overlook restriction enzyme (RE) cut site density or treat it as a technical bias requiring normalization or removal, discarding chromatin accessibility information that distinguishes functional regulatory elements. Here we introduce sintHiChIP to address this methodological gap. sintHiChIP explicitly models RE cut site density as a biological signal and integrates Gaussian kernel smoothing with distance-dependent statistics, allowing detection of chromatin loops while capturing local regulatory heterogeneity. Moreover, the algorithm employs adaptive probability distributions to resolve inherent data overdispersion and sparsity dynamically. Validation against independent datasets and comparison with existing methods demonstrate that sintHiChIP reliably recovers canonical chromatin loops, exhibiting distinct superiority in regulatory H3K27ac environments and comparable accuracy in structural cohesin contexts. Notably, sintHiChIP achieves exceptional precision in predicting CRISPRi experiments and reveals highly coherent cell-type-specific genetic regulatory networks. Executing efficiently on standard workstations, our method delivers a promising analytical framework for functional 3D genomic studies.