RefHiC

RefHiC leverages a reference panel of Hi-C contact maps and an attention-based deep learning framework to improve annotation of loops and topologically associating domains (TADs) in Hi-C data for studies of 3D genome organization and gene regulation.

Key Features:

• Loop and TAD annotation: Enhances annotation of chromatin loops and topologically associating domains (TADs) at high resolution from Hi-C data.
• Reference panel integration: Leverages a reference panel of publicly available Hi-C contact maps to mitigate limited sequencing coverage in individual datasets.
• Attention-based deep learning: Employs an attention-based deep learning framework to incorporate information from reference maps into annotations.
• Multi-dataset integration: Integrates data from multiple Hi-C datasets to improve signal detection and annotation reliability.
• Conservation exploitation: Capitalizes on the conservation of topological structures across cell types and species to inform annotations.
• Accuracy improvement: Improves accuracy and reliability of loop and TAD calls compared to methods analysing individual Hi-C datasets.
• Empirical validation: Demonstrated consistent outperformance of other tools in TAD and loop annotation across diverse cell types, species, and sequencing depths.

Scientific Applications:

• Hi-C feature annotation: Annotating chromatin loops and TADs in new Hi-C study samples.
• 3D genome organization studies: Investigating relationships between genome topology and gene regulation.
• Comparative genomics: Comparing topological structures across cell types and species using conserved features in reference panels.
• Low-coverage analyses: Enhancing structural annotation in datasets with limited sequencing depth by leveraging reference maps.

Methodology:

RefHiC applies an attention-based deep learning framework that integrates a reference panel of publicly available Hi-C contact maps and multiple Hi-C datasets, exploiting conserved topological structures across cell types and species to improve loop and TAD annotation.