RelAge-GNN

RelAge-GNN achieved the strongest chronological-age correlation among the reported models while keeping error rates competitive. RelAge-GNN is presented as better at detecting disease-associated deviations from expected aging trajectories than ordinary regression metrics alone show. Each relational graph is processed by an independent graph neural network branch and fused with a learnable node-level gate. RelAge-GNN builds three separate relational graphs over the same CpG nodes. The model predicts biological age from methylation and genomic features through node compression, graph readout, and an MLP regression head.