CRCNS Research Proposal: Collaborative Research: Discovering Network Structure in the Space of Group-Level Functional Differences

点击翻译按钮获取中文摘要

Large-scale study correlated patterns of activity in the brain (functional connectivity) can provide a unique glimpse into the inner workings of neuropsychiatric functions and disorders. However, current methods follow a less than optimal procedure for clinical analyses: they first fit a model to each individual, and then separately identify group differences. In practice, this approach tends to implicate distributed functional changes across the brain, which are difficult to interpret, ignore crucial information about the patient cohort, and fail to replicate across studies. This project takes an entirely new look at this problem by hypothesizing that each neuropsychiatric disorder reflects a set of coordinated disruptions in the brain. As a result, the induced functional differences between patients and neurotypical controls should be interdependent and form their own subnetwork. This strategy reflects a growing perception in the field that complex neuropsychiatric disorders are system-level dysfunctions, rather than collections of isolated effects. Going one step further, the inference procedures developed in this work will strategically leverage patient heterogeneity to guide the subnetwork estimation. In this end, this project will pave the way for robust and targeted biomarker discovery across a wide range of neuropsychiatric disorders.The technical exploration of this project will unfold in three stages, each of which incorporates an additional level of abstraction. Task 1 is to develop a core model of network-based functional differences via two complementary topologies. Namely, a community architecture suggests that the given deficit arises from a subset of abnormally communicating brain regions, whereas a spreading model assumes that the deficit is linked to a sparse set of region hubs, which abnormally interact with the rest of the brain. Task 2 will broaden the core framework by incorporating structural information from diffusion MRI and by estimating time-varying network differences. Finally, Task 3 will take a purely data-driven approach to the network estimation based on semi-supervised representation learning. In parallel with these technical innovations, Task 4 will address key clinical questions related to three of the most prevalent neurodevelopmental disorders: autism, ADHD, and schizophrenia. The principal investigators will release a flexible computational platform for functional connectomics based on the results of this project.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.