HiCBridge: Resolution Enhancement of Hi-C Data Using Direct Diffusion Bridge

Jaemin Kim, Jong Chul Ye

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: Hi-C, Image Translation, Diffusion Bridge, High Resolution

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: We propose HiCBridge that learns transformation from low-resolution Hi-C data to high-resolution ones using direct diffusion bridge.

Abstract: Hi-C analysis provides valuable insights into the spatial organization of chromatin, which affects many aspects of genomic processes. However, the usefulness of Hi-C is hindered by its resolution limitations. Here, we propose Hi-C enhancement using Direct Diffusion Bridge (HiCBridge) that learns transformation from low-resolution Hi-C data to high-resolution ones using direct diffusion bridge (DDB). Instead of relying on standard supervised feed-forward networks and GANs, which often produces overly smooth textures or falls into mode collapsing, the main idea of HiCBridge is building a diffusion process, by directly bridging the low and high-resolution Hi-C data. Furthermore, to make our model applicable in real-world situations, we further train our model by increasing the variation of the real-world data with diffusion model-based data augmentation. We demonstrate that our model can be used to improve downstream analyses such as three-dimensional structure matching, loop position reconstruction, and recovery of biologically significant contact domain boundaries. Experimental results confirm that HiCBridge surpasses existing deep learning-based models on standard vision metrics, and exhibits strong reproducibility in Hi-C analysis of human cells.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 3128