Collaborative Research: eMB: A deep learning-enhanced multiphysics and multiscale framework for mechanistic modeling of erythrophagocytosis

Erythrophagocytosis is a complex multiphysics process involving recognizing, engulfing, and digesting aged or diseased red blood cells (RBCs) by phagocytic cells. Biochemical signaling pathways mediated by ligand-receptor engagement have been considered as key factors in initiating and driving the phagocytosis of abnormal RBCs by tissue-resident macrophages in the spleen and the liver. However, growing evidence has underscored the effect of the stiffness of RBCs in modulating the engulfment process. Building on this evidence, the project proposes that erythrophagocytosis is not only governed by the biochemical signaling pathways but is also significantly impacted by the mechanics of RBCs. To validate the hypothesis and address the key question of how multiple biochemical signaling pathways and RBC biomechanics are intertwined in dictating the erythrophagocytosis, the project will develop an artificial intelligence (AI)-enhanced multiphysics and multiscale framework validated using multimodal experimental data. The project will apply this framework to quantify the impact of signaling pathways and RBC stiffness on macrophage-mediated RBC engulfment. The proposed framework is transformative to investigate the pathogenesis of various hemolytic anemia and the mechanisms of macrophage-based approaches for cancer immunotherapy. Integration of biochemical and biomechanical modeling using AI approaches bridges the gap between the spatial and temporal scales of molecular and cellular interaction, opening a new avenue to address a wide range of biological and biomedical questions. Research outcomes will be disseminated into three courses at three universities. The project will recruit undergraduate and high school students and actively involve them in the research. The project will develop two multiphysics models using different deep learning algorithms to perform multiscale analyses of erythrophagocytosis. In Model 1, the project will incorporate the role of RBC stiffness into the biochemical signaling model of erythrophagocytosis by adding a new pathway. This system-level model, which is suitable for making predictions across blood samples, will be built using an AI-enhanced pipeline consisting of identifiability analysis and systems biology-informed neural networks (SBINNs). While identifiability analysis is used to optimize model design, SBINNs enhance efficiency in inferring model parameters from limited experimental data. In Model 2, the project will integrate biochemical signaling models with biomechanical models to drive the multiscale process of erythrophagocytosis. This multiphysics and multiscale model enables the simulation of various subcellular processes, i.e., the formation of actin filaments and their interaction with the plasma membrane, and cellular level process including the interaction between the macrophages and their targets as well as the internalization of targets. The project will bridge the sub-cellular model and the cellular model using deep neural operators to improve the computational efficiency. Model 2 is feasible for investigating the molecular mechanisms underlying erythrophagocytosis at the single-cell level. The two proposed models will be informed and validated using data from existing and new phagocytosis experiments. In summary, the project will develop new multiphysics and multiscale models powered by deep learning to elucidate the complex interplay between biochemical signaling and biomechanics in regulating erythrophagocytosis. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.