CAREER: Multiscale Model-Driven Analysis of the Brain and its Disorders

The multiscale nature of the brain is arguably the most significant impediment to understanding its inner workings. This project aims to establish a modeling framework for addressing pressing neuroscience-related questions, and to demonstrate this capability by tackling clinically relevant research questions related to atypical neural activity in autism. This project is expected to have lasting societal benefits by developing analytical methods further empowering the study of the brain and the myriad of psychiatric, neurological, neurodevelopmental, and neurodegenerative conditions. Proposed modeling efforts will fuel synergistic ongoing collaborative projects on biologically-inspired learning systems, further sustaining the growth of artificial intelligence and deep learning. This research also integrates with educational objectives through a series of didactic videos and a one-week workshop. The brain is a tremendously complex system operating at multiple spatiotemporal scales. No framework has proved versatile and powerful enough to integrate the heterogeneous sources of information required to understand this system across scales. This project proposes a novel model-driven approach to address these gaps. Specifically, it will implement a multiscale forward model linking cellular mechanisms with whole-brain dynamics, integrating spiking neural networks and neural masses using co-simulation (Aim 1). It innovates by conceptualizing and implementing a framework informing macroscopic analyses (e.g., EEG) from microscale mechanisms constrained by mesoscale organizational principles. Further, it will leverage deep Bayesian learning for model parameter inference from simulated EEG, demonstrating that physiologically relevant latent variables can be studied from experimental data using multiscale, integrative modeling (Aim 2). The power of this model-driven approach will be illustrated using EEG recorded in genetically defined groups to investigate whether the imbalance in excitatory/inhibitory ratio in autism is associated with a breakdown in long-range effective connectivity, a use case of crucial relevance for understanding neural mechanisms and their dysregulation in autism (Aim 3). By using deep Bayesian learning for multiscale inference, this project is expected to alleviate scalability issues with traditional approaches and have a transformative effect on computational neuroscience methods and neuroscience data analysis. 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.