Learning Multi-Relational Graph Representations for DNA Methylation-Based Biological Age Estimation

Researchers introduce RelAge-GNN, a graph neural network framework that models complex biological relationships among DNA methylation sites to improve aging clock predictions. The method outperforms existing approaches in estimating biological age and shows enhanced sensitivity for detecting age acceleration in disease cohorts, with interpretability analysis revealing which relationships and CpG sites drive predictions.

RelAge-GNN addresses a fundamental limitation in computational biology: most existing aging clock models treat DNA methylation sites as independent features, ignoring the intricate web of biological relationships that govern aging processes. This research bridges that gap by constructing multiple complementary graphs—capturing co-methylation patterns, genomic proximity, and gene-level associations—and combining them through a learnable gating mechanism. The approach demonstrates measurable improvements in both predictive accuracy and correlation with chronological age across large-scale datasets.

The work emerges from a broader trend in machine learning toward structure-aware models that respect domain knowledge. In genomics specifically, graph neural networks have gained traction because they naturally encode biological networks and interactions. Previous aging clock research relied heavily on statistical methods or simple neural networks, creating opportunities for performance gains through better feature engineering and relational modeling.

The practical implications extend beyond academic interest. Accurate biological age estimation has clinical value for identifying individuals at elevated disease risk and monitoring intervention effectiveness. The model's improved sensitivity to age acceleration across disease cohorts suggests potential applications in precision medicine and drug development pipelines. Biotech companies developing therapeutic interventions for aging-related conditions could leverage such improvements in their clinical trial designs and patient stratification strategies.

Looking ahead, the interpretability findings—quantifying contributions of different relational structures and specific CpG sites—provide a foundation for mechanistic aging research. Future work might explore whether identified biomarkers become targets for therapeutic intervention or refined diagnostic criteria. The open-source availability also enables broader adoption and external validation across diverse populations.

• →RelAge-GNN integrates three complementary biological relationship graphs to model DNA methylation patterns more accurately than independent-feature approaches.
• →The framework achieves competitive accuracy with improved correlation to chronological age and enhanced sensitivity for detecting age acceleration in disease cohorts.
• →Learnable gating mechanism adaptively fuses graph representations, allowing the model to weight different relational structures based on data.
• →Post hoc interpretability analysis reveals which CpG sites and biological relationships drive predictions, supporting mechanistic insights into aging.
• →Open-source code availability enables reproducibility and adoption in biotech and clinical research applications.