Multiscale simulations that incorporate patient-specific neural network models of platelet calcium signaling predict diverse thrombotic outcomes under flow.

During thrombosis, platelets rapidly deposit and activate on the vessel wall, driving conditions such as myocardial infarction and stroke. The complexity of thrombus formation in pathological flow geometries, along with patient-specific pharmacological responses, presents an opportunity for computational modeling to help deliver novel diagnostic and therapeutic insights. In the present study, we employed a multiscale 3D computational model that incorporates unique donor-derived neural networks (NNs) trained with platelet calcium mobilization traces under combinatorial exposure to 6 agonists (n = 10 donors). The 3D model comprises four modules: a donor-specific NN model for platelet calcium mobilization, a lattice kinetic Monte Carlo solver for tracking platelet motion and bonding, a finite volume method solver for modeling soluble agonist release and convective-diffusive transport, and a lattice Boltzmann method solver for predicting the blood velocity field. Simulations were conducted for platelets from individual blood donors under venous and arterial flow conditions on a defined collagen surface, examining the effects of inhibiting ADP and TXA2, as well as the influence of nitric oxide and prostacyclin. The results reveal significant individual variability in platelet responses, influencing simulated thrombus growth dynamics and emphasizing the importance of personalized models for predicting thrombotic behavior. This approach enables consideration of patient-specific platelet signaling, drug responses, and vascular geometry for predicting thrombotic episodes, essential for advancing precision medicine and improving patient outcomes in thrombotic conditions.