Biological Pathway Informed Models with Graph Attention Networks (GAT)

Biological pathways map gene–gene interactions that govern all human processes. Despite their importance, most ML models treat genes as unstructured tokens, discarding temporal and structural priors. Recent pathway-informed methods add pathway-pathway priors, but still pool genes within a pathway using MLPs, losing mechanistic detail. We introduce a relation-aware Graph Attention Network (GAT) that performs multi-relation (e.g. activating, inhibiting) message passing over curated pathway graphs to model time-series mRNA data. We show that GATs learn temporal dynamics that generalize much better than MLPs, achieving an 81\% reduction in MSE when predicting future gene expressions under unseen conditions. We further validate the correctness of our biological prior by encoding drug mechanisms via edge interventions, boosting model robustness. Finally, trained on no prior, the GAT model correctly rediscovers all 5 signed gene-gene interactions in the canonical TP53-MDM2-MDM4 feedback loop from raw mRNA data, highlighting GATs as a powerful framework for learning causal, interpretable biological relationships from time-series experimental data. All data and code is publicly available at: https://github.com/GavinWongYF/biological-pathway-gat.