Integrating Long-Range Regulatory Interactions to Predict Gene Expression Using Graph Convolutional Networks.

Area of Science:

• Genomics
• Computational Biology
• Epigenetics

Background:

• Gene expression is regulated by long-range genomic interactions, crucial for cellular function.
• Disruption of these interactions is linked to various diseases.
• Current deep learning models often overlook 3D genomic structure and long-range interactions.

Purpose of the Study:

• To develop a novel deep learning model that integrates 3D genomic organization and regulatory factors for gene expression prediction.
• To capture long-range regulatory interactions and their impact on gene expression.
• To provide an interpretable model that identifies key regulatory factors and genomic regions.

Main Methods:

• Developed a Graph Convolutional Model for Epigenetic Regulation of Gene Expression (GC-MERGE).
• Utilized a graph-based framework to encode spatial genomic interactions.
• Integrated multimodal data, including global genomic organization and local histone modifications.

Main Results:

• GC-MERGE demonstrated state-of-the-art predictive performance on GM12878, K562, and HUVEC cell lines.
• The model successfully predicted gene expression by incorporating 3D genomic structure and epigenetic data.
• Model interpretability allowed identification of significant histone modifications and interacting genomic regions.

Conclusions:

• GC-MERGE offers a novel approach for predicting gene expression by integrating multimodal data within a graph convolutional framework.
• The model provides interpretable insights into the biological mechanisms underlying gene regulation.
• This approach enhances understanding of disease-associated regulatory disruptions.