Data-driven multiscale modeling and physics-informed machine learning methods
Alireza Yazdani, PhD
I will first present a novel physics-informed deep learning framework, where Navier-Stokes informed neural networks that encode the governing equations of fluid motions i.e., mass, momentum and transport equations are used to infer hidden quantities of interest such as velocity and pressure fields merely from spatio-temporal visualizations of a passive scaler (e.g., dye or smoke) transported in arbitrarily complex domains. This inverse approach can be used for physical and biomedical problems to extract valuable quantitative information (e.g., lift and drag forces or wall shear stresses in arteries) for which direct measurements may not be available. Next, I will discuss recent developments in computational modeling of complex biological systems at both meso- and macroscopic scales and show the performance of our modeling strategies in addressing thrombus (clot) formation in aortic dissections — a life-threatening event that is initiated by damage in arterial wall propagating within the media layer and connecting with the true lumen to form a so-called false lumen within the aortic wall — by using in-vivo and in-vitro data collected for murine dissections. At the continuum level, we solve the Navier-Stokes equations for blood flow along with the advection-diffusion-reaction (ADR) equations for multiple species in the coagulation cascade. To model platelet aggregation, we employ the force coupling method (FCM) that provides a flexible platform for two-way coupling of platelets with the background flow, where the thrombus modeled by FCM is affected by the local hydrodynamics and fluid stresses. At cellular and sub-cellular levels, however, particle-based methods such as Dissipative Particle Dynamics (DPD) have proven themselves to be more effective in addressing complex biological systems. The connections between these two frameworks and future directions will also be discussed.