Integrative Statistical Models of Large-scale Brain Networks

Loading...
Thumbnail Image

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

Modern neuroscience is producing neural population activity datasets of unprecedented scale. Despite this, many analyses of these data continue to rely on underspecified null models that do not incorporate established principles of brain organization. Such limitations make it difficult to distinguish genuinely new neuroscientific discoveries from new interpretations that simply reflect known structure and thus do not advance our understanding of brain function in both health and disease. These challenges underscore the need to develop benchmark models that formalize existing knowledge and provide principled baselines against proposed discoveries. This dissertation develops a unified mathematical and computational framework for constructing such benchmark models using nullspace sampling. Nullspace sampling is a statistical method that generates synthetic neural population activity that satisfies specified statistical constraints while remaining maximally random in all other respects. A central contribution of this work is a scalable hierarchical framework grounded in multiscale modularity, an evolutionarily conserved organizational principle of neural systems. When applied to whole-brain zebrafish and cortex-wide mouse recordings, this framework demonstrates that hierarchical modularity alone can reproduce key empirical signatures of large-scale population activity, including correlation structure, tuning organization, and nonlinear low-dimensional geometry. These findings position hierarchical modularity as a parsimonious scaffold capable of explaining diverse aspects of neural dynamics at scale. In addition, to illustrate the broader utility of benchmark-model analysis, this work also applies nullspace sampling methodology to recordings from primate working memory. The work indicates that the framework enables controlled comparisons that clarify how distinct hypothesized components of population activity contribute to stimulus-related decoding, highlighting how benchmark models can help adjudicate theoretical proposals across domains. Overall, this work advances a principled set of tools for evaluating neuroscientific results against strong baselines and thus accelerating fundamental research across systems and network neuroscience.

Description

Keywords

Network Neuroscience, Computational Neuroscience, Statistical Modeling, Hierarchical Modularity, Working memory

Citation

Endorsement

Review

Supplemented By

Referenced By