A Region-Aware Structured Framework Improves Prediction of Gene Expression from DNA Methylation.

Area of Science:

• Epigenetics
• Computational Biology
• Genomics

Background:

• DNA methylation is a crucial epigenetic mark regulating gene expression and implicated in disease.
• Predicting gene expression from DNA methylation enables cross-omics integration and understanding of regulatory mechanisms.
• Current methods struggle to capture complex, region-specific, and nonlinear methylation-gene expression interactions.

Purpose of the Study:

• To develop a novel neural network framework, RSMethy-Net, for accurate gene expression prediction from DNA methylation data.
• To address the limitations of existing methods in modeling region-aware nonlinear relationships between DNA methylation and gene expression.
• To provide insights into the distinct regulatory patterns of methylation across different gene functional regions.

Main Methods:

• Proposed RSMethy-Net, a neural network incorporating grouped region encoding modules for distinct gene functional regions.
• Developed a nonlinear predictive framework to capture latent regulatory patterns and methylation-expression associations.
• Systematically evaluated model performance across 6 cancer cohorts, comparing against multiple baseline methods.

Main Results:

• RSMethy-Net demonstrated superior predictive performance compared to existing baseline methods.
• The region-aware design allowed for quantification of contributions from different gene regions to expression prediction.
• Experimental results validated the model's effectiveness in capturing complex methylation-expression dynamics.

Conclusions:

• RSMethy-Net offers an effective approach for predicting gene expression from DNA methylation, accounting for region-specific nonlinearities.
• The framework provides valuable insights into the functional roles of DNA methylation in different genomic regions.
• This study advances computational strategies for epigenetic data analysis and understanding gene regulation in disease contexts.