Learning the Dynamics of Biological Networks with Deep Graph Networks

Large scale biological networks have expanded significantly due to advancements in high-throughput technologies, yet they remain static representations of complex biological processes. In contrast, Biochemical Pathways (BPs) model cellular processes dynamically, providing insights through numerical simulations.

This work explores the potential of studying dynamic properties directly on Protein-Protein Interaction Networks (PPINs) using a neural model from the Deep Graph Network (DGN) family, with a focus on predicting sensitivity—how changes in one protein’s activity affect another. We simulate sensitivities from BPs, mapping them onto PPINs, and leverage DGN models to infer these relationships from topological features.

The approach aims to bypass the need for detailed biochemical models, providing scalable methods for applications in drug discovery, repurposing, and personalized medicine.