CRII: RI: Using Large-Scale Neuroanatomy Datasets to Quantify the Mesoscale Architecture of the Brain

揭示大脑全局连接的方法通常是从一次追踪少量神经元开始的。正是通过在不同的大脑区域和许多大脑中进行许多实验，信息才能被聚合和整合，以产生大脑全球网络和架构的详细地图。这个项目的目的是开发新的计算方法来模拟小鼠大脑的连通性，以揭示连接和信息路由的原理。该项目将利用来自艾伦脑科学研究所的全脑成像数据集，每个数据集都提供了拼图的一小部分，但结合在一起，可以得出全脑连接的图像。这项研究的结果将是小鼠大脑全球连接的新地图，以及研究疾病和衰老对全脑网络影响的框架。该项目将开发一个新的框架，用于分析全脑连接数据集，以模拟高层(中观)的布线和架构原理。为了做到这一点，矩阵分解的工具将被用来将来自许多大脑的大型数据集分解成一组已学习的神经路径或“部分”，当组合在一起时，尽可能简洁地描述大量数据。为了解决数据集的大小，将探索基于次抽样的方法和随机方法的使用，以用于大规模的机器学习应用。通过对数据进行智能和自适应的二次抽样，结合因式分解的在线方法，将制定方法，以低于10微米的分辨率处理和学习整个艾伦研究所的老鼠连接图集。该项目的成果将是大规模矩阵分解的新工具，全脑连接体的模型，以及可能有助于开发下一代机器智能体系结构的布线原理的发现。该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

Methods for revealing the global connections of the brain typically start by tracing a small number of neurons at a time. It is through performing many experiments, in different brain areas, and across many brains, that information can be aggregated and consolidated to produce detailed maps of the brain's global networks and architecture. The aim of this project is to develop new computational approaches for modeling the connectivity of the mouse brain, in order to reveal principles of wiring and information routing. The project will leverage whole-brain imaging datasets from the Allen Institute for Brain Science that each provide a small piece of the puzzle but when combined, can yield a picture of whole-brain connectivity. The outcomes of this research will be new maps of the global connectivity of the mouse brain, and a framework for studying the impact of disease and aging on whole-brain networks.This project will develop a novel framework for analyzing whole-brain connectomics datasets to model high-level (mesoscopic) principles of wiring and architecture. To do this, tools from matrix factorization will be used to decompose large datasets from many brains into a collection of learned neural pathways or "parts" that, when combined, describe large volumes of data as succinctly as possible. To address the size of the datasets, the use of subsampling-based approaches and randomized methods will be explored for massive-scale machine learning applications. Through intelligent and adaptive subsampling of the data, in combination with online methods for factorization, methods will be developed to process and learn from the entire Allen Institute Mouse Connectivity Atlas at less than 10-micron resolution. The outcomes of this project will be new tools for large-scale matrix factorization, models of the whole brain connectome, and discovery of wiring principles that could be useful in the development of next generation architectures for machine intelligence.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.