HiCMamba: Enhancing Hi-C resolution and identifying 3D genome structures with state space modeling

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• High-throughput chromosome conformation capture (Hi-C) is crucial for studying 3D genome organization.
• Limited sequencing coverage in Hi-C data leads to imprecise chromatin interaction frequency estimates.
• Existing methods struggle with computational efficiency and resolution enhancement.

Purpose of the Study:

• To develop a novel deep learning method, HiCMamba, for enhancing Hi-C contact map resolution.
• To address the limitations of low-coverage Hi-C data.
• To improve the accuracy of 3D genome structure identification.

Main Methods:

• Utilized a UNet-based auto-encoder architecture incorporating a holistic scan block.
• Employed a state space model for Hi-C resolution enhancement.
• Developed HiCMamba, a deep learning approach for processing Hi-C data.

Main Results:

• HiCMamba significantly outperforms existing state-of-the-art methods in Hi-C resolution enhancement.
• The method achieves superior results while demanding fewer computational resources.
• 3D genome structures (TADs, loops) identified by HiCMamba are validated by epigenomic features.

Conclusions:

• State space models show significant potential as foundational frameworks for Hi-C resolution enhancement.
• HiCMamba offers a computationally efficient and effective solution for improving Hi-C data resolution.
• The developed method facilitates more accurate analysis of 3D genome structure and function.