ReHiC: Enhancing Hi-C data resolution via residual convolutional network.

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• High-throughput chromosome conformation capture (Hi-C) is crucial for studying genome 3D organization.
• Current Hi-C methods are costly, time-consuming, and often yield low-resolution data insufficient for small-scale pattern identification.

Purpose of the Study:

• To introduce ReHiC, a deep learning computational approach for enhancing Hi-C data resolution.
• To enable high-resolution Hi-C data generation at a reduced cost.

Main Methods:

• Developed ReHiC, a deep learning model utilizing residual blocks for faster convergence and deeper networks.
• The model requires only a 1/16 down-sampling ratio of original Hi-C data to predict higher resolution matrices.
• Compared ReHiC performance against established methods like HiCPlus and HiCNN.

Main Results:

• ReHiC accurately predicts high-resolution Hi-C data, closely matching numerical and interaction distributions of true high-resolution data.
• The model demonstrates superior performance compared to HiCPlus and HiCNN.
• The ReHiC framework shows portability, effectively enhancing Hi-C matrices from different cell types.

Conclusions:

• ReHiC provides a cost-effective solution for generating high-resolution Hi-C data.
• The method offers more accurate spatial organization insights by enabling detailed genomic interaction analysis.
• ReHiC represents a significant advancement in computational approaches for genomic structure analysis.