Multiscale theory of synapse function with model reduction by machine learning

点击翻译按钮获取中文摘要

Project Summary/Abstract This project constructs a unifying model that links synaptic morphodynamics, the fundamental process of learning and memory in the brain, to the underlying molecular signaling pathways that regulate it. The motivation for this work is a new class of machine learning methods for multiscale modeling that are a promising candidate for linking the disparate spatial and temporal scales involved, from s calcium events in nano-domains to actin reorganization on the order of minutes across a dendritic spine head. Previously, it has only been possible to study each of these scales in isolation. The project brings together experts in (1) modeling the biochemistry at synapses, (2) modeling the growth of the actin cytoskeleton, and (3) developing the theory and algorithms of multiscale modeling with machine learning. The result of this collaboration will be a milestone model in cellular neuroscience that mechanistically connects calcium signaling in dendritic spines to the growth of the actin cytoskeleton in spine remodeling. Currently, there are few models that can e ectively make predictions about actin structure formation based on changes in calcium in ux into the post-synaptic spine. Since the new data-driven models will be more computationally ecient than exact simulations, it will also be possible to incorporate them into coarse-scale models of synapses used in network simulations and in neuroengineering applications. Additionally, the methods developed in this work an important contribution to modeling in cellular neuroscience, particularly because they are data-driven and therefore widely applicable. Finally, the development of a suite of software tools for multiscale modeling with machine learning will catalyze future collaborations and scienti c developments in the neuroscience community, particularly using models that aim to connect cellular phenomena with mechanisms at sub-second resolution. Such models can potentially bene t the development of pharmaceutical targets for learning de cits associated with aging and neurological disorders such as Alzheimers.