Integrative chromatin domain annotation through graph embedding of Hi-C data

Area of Science:

• Genomics
• Computational Biology
• Epigenetics

Background:

• Genomic domain organization is crucial for gene expression and cellular functions.
• Current methods for identifying genomic domains rely on 1D functional assays (e.g., ChIP-seq) or 3D chromatin conformation assays (e.g., Hi-C).
• Existing integrative modeling approaches often use segmentation and genome annotation (SAGA) with a potentially flawed assumption that physically interacting loci share the same domain type.

Purpose of the Study:

• To develop an integrative computational method for annotating genomic domains.
• To combine 1D functional genomic signals and 3D Hi-C interaction data without relying on pairwise priors.
• To improve the granularity and accuracy of genomic domain identification.

Main Methods:

• Developed an integrative approach utilizing graph embedding to learn structural genomic region features.
• Integrated learned structural features with 1D functional genomic signals.
• Applied a segmentation and genome annotation (SAGA) algorithm to the combined data.

Main Results:

• The new domain types accurately recapitulate known genomic subcompartments.
• The approach provides enhanced granularity, distinguishing combined spatial and functional states of genomic regions.
• A specific division within the A2 subcompartment was identified, showing significantly different expression levels.

Conclusions:

• The developed integrative method offers a more refined annotation of genomic domains.
• This enhanced domain resolution provides deeper insights into the relationship between 3D genome organization and gene expression.
• The approach successfully integrates diverse genomic data types for improved domain discovery.