PB-DiffHiC: a statistical framework for detecting differential chromatin interactions from high resolution pseudo-bulk Hi-C data.

Single-cell Hi-C (scHi-C) data provide unprecedented opportunities for analyzing differential chromatin interactions, essential for understanding genome structure-function relationships across various biological conditions. However, biologically meaningful differential chromatin interaction analysis at high resolution (e.g., 10 Kb) remains challenging due to the inherent sparsity of scHi-C data. Existing approaches typically rely on single cell imputation, which is computationally intensive and lacks validation, or apply conventional bulk Hi-C tools to pseudo-bulk matrices aggregated from individual cells. The sparsity of high-resolution pseudo-bulk data limits the effectiveness of bulk-oriented methods. Here, we present PB-DiffHiC, an optimized parametric statistical framework that directly analyzes raw pseudo-bulk Hi-C data at 10 Kb resolution between conditions. PB-DiffHiC incorporates Gaussian convolution, the stability of short-range interactions, and Poisson modeling to jointly perform normalization and statistical testing. Benchmarking on cell-type-specific chromatin loops shows that PB-DiffHiC achieves higher precision than alternative methods. Application to pseudo-bulk and matched bulk Hi-C data demonstrates stronger concordance in identified differential interactions, reinforcing its reliability. In a case study, PB-DiffHiC successfully identifies Kcnq5-associated differential interactions that closely matching SnapHiC-D results, despite not relying on single-cell imputation. PB-DiffHiC is a statistically sound and robust method for high-resolution differential analysis of chromatin interactions using raw pseudo-bulk Hi-C data. The source code of PB-DiffHiC is publicly available at https://github.com/Tian-Dechao/PB-DiffHiC .