Speaker
Description
Computational models provide a biophysically grounded framework to study tau protein spread in Alzheimer's disease and other tauopathies. However, model parameter inference, a key step in analyzing empirical data, is often intractable due to the slow speed and high computational burden of generating simulations.
Here, we introduce a machine learning framework to accelerate the Network Transport Model (NTM), which models tau spread on the brain's structural connectome graph. We replaced the computationally expensive graph edge partial differential equation (PDE) solver subcomponent with three classes of surrogate models: linear regression, symbolic regression, and multilayer perceptrons (MLP). The MLP approach achieved the highest fidelity, attaining $R^2 > 0.99$ against ground-truth numerical solvers, while symbolic regression yielded nontrivial analytical relationships between tau spread and kinetic NTM parameters. Crucially, these approaches reduced simulation time from hours to minutes (approx 60x speed-up), enabling the thousands of forward simulations required for data-driven parameter inference.
