DiCARN-DNase: enhancing cell-to-cell Hi-C resolution using dilated cascading ResNet with self-attention and DNase-seq...

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Chromatin's spatial organization is crucial for gene regulation and cellular function.
• High-resolution Hi-C data is essential for 3D genome analysis but often limited.
• Existing deep learning methods for Hi-C resolution enhancement face challenges in detail preservation and cross-cell line generalization.

Purpose of the Study:

• To develop a deep learning model for improving Hi-C data resolution.
• To address limitations of existing methods, including blurring and poor generalization across cell types.
• To enhance the analysis of 3D genome structures by enabling high-resolution data reconstruction.

Main Methods:

• Proposed DiCARN (Dilated Cascading Residual Network) model.
• Utilized dilated convolutions and cascading residuals to capture broader genomic context and fine-grained interactions.
• Integrated DNase-seq data to improve model robustness and generalizability.

Main Results:

• DiCARN effectively enhances Hi-C data resolution.
• The model preserves fine-grained genomic interactions, overcoming blurring issues.
• Demonstrated superior generalizability across different cell lines for Hi-C data reconstruction.

Conclusions:

• DiCARN offers a robust framework for high-resolution Hi-C data reconstruction.
• The integration of DNase-seq data significantly improves cross-cell line generalization.
• This advancement facilitates more comprehensive analysis of 3D genome organization and gene regulation.