A deep learning-enhanced multiphysics and multiscale framework for predictive modeling of inflammation-induced thrombosis - As witnessed in Coronavirus disease 2019 (COVID-19) pandemic, inflammation-induced thrombosis could lead to elevated disease severity, worsened clinical outcomes and increased mortality. In particular, accumulating evidence from autopsies suggests that these thrombi are prevalent within the microvascular of vital organs, such as the lungs, liver, kidney and heart, implying that microthrombosis may contribute to the microvascular dysfunction and multi-organ failure in patients with severe cases of COVID-19. Although many proinflammatory and prothrombotic factors associated with COVID-19 have been reported, how these factors regulate the thrombosis in the microvasculature is elusive. Since the formation of microthrombi cannot be observed directly in vivo, these is a lack of markers to assess the extent of thrombogenesis and connect it to the clinical outcomes in time for appropriate therapeutic intervention. In this application, we propose to develop a deep learning-enhanced multiphysics and multiscale framework for predictive modeling of inflammation-induced thrombosis in infectious diseases. Using COVID-19 as an initial disease model, we will examine the feasibility of using circulating cell clusters (CCCs) in the blood samples of patients with COVID-19 as potential markers for disease prognosis. We will also investigate the underlying mechanisms of the formation of microthrombi and CCCs as well as predict their adverse effects in microcirculation. We propose to achieve our objectives through the following three Aims. In Aim 1, we will develop a new deep learning model based on graph neural networks to automatically analyze the flow cytometry images of CCCs from COVID-19 blood samples and explore the association between CCC phenotypes and the specific clinical outcomes of patients with COVID-19. In Aim 2, we will develop new deep learning models for coagulation pathways based on system-biology informed neural networks to explore the mechanism for the activation of inflammation-induced coagulation. In Aim 3, we will develop new multiphysics and multiscale models to simulate the formation of microthrombi and CCCs as well as their dynamics in the microvasculature. The clinical data collected in Aim 1 will be used to inform the computational models in Aims2&3 whereas Aims2&3 will provide mechanistic rationale for findings in Aim 1 and offer insight for new treatment strategies. In summary, the proposed predictive multiphysics and multiscale computational framework will provide new computational and modeling tools to improve disease prognosis for COVID-19, elucidate the pathogenesis of inflammation-induced thrombosis, identify the potential key factors that regulate the thrombogenesis, thereby opening new avenues for developing more effective and tailored antithrombotic treatments for hyperinflammatory prothrombotic disorders in various infectious diseases.
