Enhancing Hi-C data resolution with deep convolutional neural network HiCPlus.

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Epigenetics

Background:

• Hi-C technology is crucial for studying 3D genome organization.
• Current Hi-C datasets often lack the resolution to link distal regulatory elements to target genes due to sequencing costs.
• High-resolution 3D genome data is essential for understanding gene regulation.

Purpose of the Study:

• To develop a computational method, HiCPlus, for inferring high-resolution Hi-C interaction matrices from low-resolution data.
• To improve the ability to link distal regulatory elements with their target genes.
• To reduce the reliance on extensive sequencing for high-resolution Hi-C data.

Main Methods:

• Development of HiCPlus, a deep convolutional neural network approach.
• Training and validation using existing low-resolution Hi-C datasets.
• Testing the model's ability to impute high-resolution interaction matrices.

Main Results:

• HiCPlus successfully imputes Hi-C interaction matrices with high similarity to original high-resolution data.
• The method achieves this using only 1/16 of the original sequencing reads.
• Models trained on one cell type demonstrate applicability to other cell or tissue types.

Conclusions:

• HiCPlus provides a computational framework to significantly enhance Hi-C data resolution.
• This advancement facilitates the study of 3D chromatin interactions and gene regulation.
• The findings reveal underlying features of 3D chromatin interaction formation.