A Chromatin-Structure-Guided Framework for Predictive and Interpretable Regulatory Genomics.

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Gene regulation is influenced by 3D chromatin organization, but linear genome models are insufficient.
• Existing Hi-C contact maps are averaged, noisy, and obscure specific biological interactions.

Purpose of the Study:

• To develop a framework (CHROME) for identifying physically specific chromatin contacts from Hi-C data.
• To integrate these contacts into graph representations for improved genomic predictions.
• To enhance the interpretability of gene regulation and variant effect modeling.

Main Methods:

• Utilized a self-avoiding polymer ensemble null model to identify non-random Hi-C contacts.
• Developed a graph attention architecture integrating sequence, accessibility, and pre-trained embeddings.
• Applied the framework to predict cell line-specific ChIP-seq profiles, eQTLs, and variant pathogenicity.

Main Results:

• CHROME successfully identified physically specific chromatin interactions.
• The framework outperformed local encoder baselines in predicting ChIP-seq profiles, generalizing to new cell lines.
• Graph embeddings improved predictions for eQTLs and variant pathogenicity, surpassing sequence-based methods.
• Attention mechanisms provided interpretability into multi-megabase regulatory influences.

Conclusions:

• Incorporating physically validated chromatin interactions significantly enhances predictive accuracy and interpretability in genomics.
• CHROME offers a novel approach to modeling gene regulation and variant effects by leveraging 3D genome structure.
• This work bridges the gap between genome conformation and functional genomic predictions.