DFHiC: a dilated full convolution model to enhance the resolution of Hi-C data

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Hi-C technology is crucial for studying genome 3D structure by mapping all paired interactions.
• High-resolution Hi-C data are essential for detailed genome structure analysis but are costly.
• Existing low-resolution Hi-C data limit the fineness of genome structure studies.

Purpose of the Study:

• To develop an effective computational method for enhancing low-resolution Hi-C data to high-resolution.
• To improve the accuracy and reliability of genome 3D structure analysis from available Hi-C datasets.
• To provide a cost-effective solution for obtaining high-resolution Hi-C data.

Main Methods:

• Proposed DFHiC, a novel method utilizing dilated convolutional neural networks.
• DFHiC generates high-resolution Hi-C matrices from low-resolution inputs.
• Dilated convolution enables exploration of global genomic patterns over longer distances.

Main Results:

• DFHiC reliably and accurately improves Hi-C matrix resolution.
• Enhanced super-resolution Hi-C data closely matches real high-resolution data.
• DFHiC excels in identifying significant chromatin interactions and topologically associating domains.

Conclusions:

• DFHiC offers a significant advancement in Hi-C data processing.
• The method provides a valuable tool for researchers studying genome 3D organization.
• DFHiC enhances the utility of existing low-resolution Hi-C datasets for biological discovery.