Combining data-driven neural networks of platelet signalling with large-scale ODE models of coagulation

Abstract

Platelet signalling during blood coagulation is conveniently modelled using data-driven neural networks since mechanistic Ordinary Differential Equation (ODE)-based models, while available, are unwieldy due to the large number of equations. On the other hand, the extra-cellular protease cascade of reactions that occur during blood coagulation are conveniently modelled using mechanistic ODE-based models. It is essential to integrate platelet signalling with the extra-cellular reaction cascade to get a representative model for in-vitro coagulation. In this paper, a neural-network-based platelet calcium calculator is combined with a large-scale ODE model for thrombin production. The combined model is used to test the effect of platelet modulators on thrombin production during in-vitro coagulation. The model predictions generate two hypotheses on relative importance of platelet modulators: they need to be compared to experimental data to confirm the same. The novelty of the study lies in the combination of existing approaches to study coagulation, and in capturing the role of platelet modulators on thrombin production: excessively large number of equations are avoided, which would be the case in a purely ODE-based approach.