Brooks Skylar J, Stamoulis Catherine
Boston Children's Hospital, Department of Pediatrics, Boston, MA, USA.
University of California Berkeley, Helen Wills Neuroscience Institute, Berkeley, CA, USA.
IEEE Int Conf Acoust Speech Signal Proc Workshops. 2023 Jun;2023. doi: 10.1109/icasspw59220.2023.10193544. Epub 2023 Aug 2.
Functional interactions and anatomic connections between brain regions form the connectome. Its mathematical representation in terms of a graph reflects the inherent neuroanatomical organization into structures and regions (nodes) that are interconnected through neural fiber tracts and/or interact functionally (edges). Without knowledge of the ground truth topology of the connectome, functional (directional or nondirectional) graphs represent estimates of signal correlations, from which underlying mechanisms and processes, such as development and aging, or neuropathologies, are difficult to unravel. Biologically meaningful simulations using synthetic graphs with controllable parameters can complement real data analyses and provide critical insights into mechanisms underlying the organization of the connectome. Generative models can be highly valuable tools for creating large datasets of synthetic graphs with known topological characteristics. However, for these graphs to be meaningful, the variation of model parameters needs to be driven by real data. This paper presents a novel, data-driven approach for tuning the parameters of the generative Lancichinetti-Fortunato-Radicchi (LFR) model, using a large dataset of connectomes (n = 5566) estimated from resting-state fMRI from early adolescents in the historically large Adolescent Brain Cognitive Development Study (ABCD). It also presents an application, i.e., simulations using the LFR, to generate large datasets of synthetic graphs representing brains at different stages of neural maturation, and gain insights into developmental changes in their topological organization.
脑区之间的功能相互作用和解剖连接构成了连接组。其以图形形式的数学表示反映了固有的神经解剖学组织，即通过神经纤维束相互连接和/或在功能上相互作用的结构和区域（节点）。在不了解连接组真实拓扑结构的情况下，功能（定向或非定向）图表示信号相关性的估计值，从中很难揭示诸如发育、衰老或神经病理学等潜在机制和过程。使用具有可控参数的合成图进行生物学意义上的模拟可以补充真实数据分析，并为连接组组织的潜在机制提供关键见解。生成模型可能是创建具有已知拓扑特征的合成图大型数据集的极有价值的工具。然而，要使这些图有意义，模型参数的变化需要由真实数据驱动。本文提出了一种新颖的数据驱动方法，用于调整生成式兰奇基尼蒂 - 福尔图纳托 - 拉迪基（LFR）模型的参数，该方法使用了来自具有历史意义的大型青少年大脑认知发展研究（ABCD）中青少年静息态功能磁共振成像估计的连接组大型数据集（n = 5566）。本文还展示了一个应用，即使用LFR进行模拟，以生成代表神经成熟不同阶段大脑的合成图大型数据集，并深入了解其拓扑组织的发育变化。
