DeepLoop enables robust mapping of DNA loops from low-depth single cell or allele-resolved Hi-C data at high resolution

High-resolution mapping of the gene regulatory chromatin interactions from noisy Hi-C contact heatmaps remains a major challenge in 3D genome research due to the complex bias structure and severe data sparsity. Here we present DeepLoop, which addresses this problem by combining a rigorous bias correction strategy (HiCorr) and deep learning-based signal enhancement techniques. We demonstrated that DeepLoop can robustly identify high-resolution chromatin interactions from low-depth and single cell Hi-C data. With DeepLoop, we for the first-time mapped the genetic and epigenetic determinants of human homolog-specific chromatin loops in GM12878 cells at 5kb-resolution. We nominated two new imprinting loci with unbalanced DNA folding in additional to the well-known H19/IGF2 locus. Interestingly, we also found that the sub-TAD loops at the DXZ4 “megadomain” boundary escapes X-inactivation, but the loops at the FIRRE “superloop” locus do not escape. Unexpectedly, DeepLoop detected four previously unknown Mb-sized heterozygous structure variants in the GM12878 genome, including two large inversions causing allele-specific gene expression via enhancer rewiring. Finally, DeepLoop also pinpointed dozens of loop regulatory SNPs that often have transcriptional consequences. Taken together, DeepLoop expands the use of Hi-C to provide genome-wide insights into the genetics of 3D genome at loop resolution. Overall design: DeepLoop robustly identify high-resolution chromatin loops regardless of noise and depth. Each text file (300K) contains a chromatin loop calling list from Hi-C data for one cell type.