RefHiC-SR

RefHiC-SR enhances Hi-C contact map resolution by applying an attention-based deep learning framework guided by a reference panel of publicly available Hi-C contact maps to improve estimation of chromatin interaction frequencies.

Key Features:

• Attention-based deep learning: Uses an attention-based deep learning model for super-resolution inference of Hi-C contact maps.
• Reference panel guidance: Leverages a reference panel composed of multiple, several hundred, publicly available Hi-C contact maps to inform inference.
• Cross-cell-type conservation: Exploits conservation of local spatial genome organization across cell types to enhance reconstruction.
• Super-resolution inference: Increases effective resolution of low-coverage Hi-C datasets through learned enhancement.
• Improved interaction estimation: Enhances signal quality and accuracy of chromatin interaction frequency estimates.
• Structural feature mapping: Enables higher-accuracy mapping of chromatin loops and topologically associating domains (TADs).
• Robustness across conditions: Demonstrates improved performance relative to methods without reference samples across diverse cell types and sequencing depths.

Scientific Applications:

• 3D genome organization: Refines Hi-C maps for analysis of three-dimensional genome architecture.
• Gene regulation: Improves identification of regulatory chromatin contacts relevant to gene regulation studies.
• Loop and TAD analysis: Enhances detection and characterization of chromatin loops and TADs across cell types and sequencing depths.

Methodology:

Applies an attention-based deep learning model that incorporates a reference panel of multiple (several hundred) publicly available Hi-C contact maps to guide super-resolution inference of a target sample's Hi-C contact map.