CAREER: Towards a Living Neuron Twin for Improving Human Cognitive Health

点击翻译按钮获取中文摘要

Alzheimer's disease is a fatal and devastating cognitive disorder that affects millions of people worldwide, posing significant challenges to global health. Despite striking efforts, there is currently no effective treatment. The overwhelming societal burden threatens our future. This project aims to develop Neuron Twin, a digital system that simulates the human brain as a dynamic system using multimodal data analysis and multidomain knowledge integration to provide an accurate and efficient prediction of Alzheimer’s disease, and ultimately elucidate a mechanistic understanding of cognitive decline. Such an innovative system will offer new insights into treatment strategies and precision medicine that can benefit the Alzheimer's disease community and broader applications of neurodegenerative diseases. Furthermore, it leverages modeling and machine learning techniques to solve complex health data science problems, discovering relationships within large datasets and overcoming barriers across different domains. The interdisciplinary effort promotes education, diversity, and collaboration by transforming research findings into instructional materials, providing training opportunities for students from diverse backgrounds, and engaging undergraduate and underrepresented students in summer bootcamp and research activities.This project focuses on developing a computational framework for the Neuron Twin system. The backbone of Neuron Twin is the coalition of deep learning and multiscale modeling, which complement each other to overcome inherent limitations and leverage method scalability. Unlike existing approaches that rely on statistical inference, this system jointly analyzes multimodal data, including genetic data, neuroimages, and clinical data, and integrates multidomain knowledge from bioinformatics, systems biology, and network neuroscience to facilitate reliable early diagnosis and prognosis of Alzheimer's disease. The framework consists of three research thrusts. The first thrust is to build a multiscale model that can capture the spatiotemporal dynamics of disease progression by synthesizing information from gene regulation, protein interaction, and phenotypic heterogeneity. The second thrust is to develop continual model-guided learning to provide neurologically consistent predictions for small data regimes and continuously improve the system with sporadic data updates. The third thrust is to design hybrid learning-aided inference to address model incompleteness in parameterization and hypothesis validation. The project will be evaluated through large-scale neuroimaging genetic studies of neurodegenerative diseases.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.