The field of biological and graph neural networks is rapidly advancing, with a focus on incorporating geometric and temporal information to improve model performance and interpretability. Recent developments have seen the proposal of novel frameworks that integrate geometric and temporal components, such as geometry-aware spiking graph neural networks and stage-aware mixture of experts frameworks. These approaches have shown significant improvements in tasks such as prognostication, disease progression modeling, and molecular representation learning. Notably, the use of geometric and temporal information has enabled the development of more accurate and robust models, such as those that can predict biological age and its longitudinal drivers.

Some noteworthy papers in this area include: Self-Organizing Survival Manifolds, which proposes a theory for unsupervised discovery of prognostic structures in biological systems by modeling survival as a geometric consequence of the curvature and flow inherent in biological state space. Geometry-Aware Spiking Graph Neural Network, which introduces a novel spiking graph neural network that unifies spike-based neural dynamics with adaptive representation learning on Riemannian manifolds, achieving superior accuracy and robustness compared to existing models.