Graph Learning Models for Temporal Gene Expression Prediction and the Role of Interactions Topology

Predicting gene expression dynamics is challenging due to the complex regulatory interactions within high-dimensional datasets. We evaluate predictive models that integrate temporal patterns with gene–gene networks, comparing a state-of-the-art approach based on Protein-Protein Interaction (PPI) networks from STRING with models utilizing data-driven network inference. Our results show that inferred networks can enhance accuracy over static biological priors. However, simpler models treating genes independently often achieve comparable performance. This suggests that for the considered datasets, the added complexity of explicit gene–gene interactions does not always translate into superior predictive power, opening to further investigations on the most effective ways to represent and leverage biological connectivity in forecasting tasks.

Recommended citation: A. Dipalma, A. Micheli, P. Milazzo, F. Simonetti, D. Tortorella (2027). "Graph Learning Models for Temporal Gene Expression Prediction and the Role of Interactions Topology." Computational Methods in Systems Biology. CMSB 2026, LNCS vol. 16678, pp. 334-350.
Read Paper