Functional Data Analysis for High-Dimensional Biobehavioral Data

项目总结 大约每59名儿童中就有1名被诊断出患有自闭症谱系障碍(ASD)，即 神经发育障碍的特征是社会交往障碍和 沟通。我们在这笔赠款中的建议是基于对两个最重要的 ASD、脑电(EEG)和眼科等生物行为标志物的前景看好 跟踪(ET)。这两项研究都是从连续实施的EEG和ET任务中收集数据的 多次纵向访问。此外，医疗设备内部或跨医疗设备的多个任务利用类似的 认知域。因此，即使对这些复杂数据结构的联合分析 任务、模式(EEG和ET)和纵向访问将导致最有效地利用 可用的信息，目前的分析技术是有限的，通常是在数据上进行的 一次一个任务，在一个通道内。因此，我们提出了一套全面的 生物行为生物标志物数据整体分析的统计方法 来自多个任务、跨医疗模式和纵向访问的信息。我们的建议 依赖于将EEG和ET数据表征为高维、高度结构化的泛函 物体。与现有的多模式脑成像文献不同，多模式脑成像文献为大脑融合数据- 与区域相关的推断，我们结合了脑成像模式(EEG)和生物行为 标记(ET)，基于与共同认知领域相关的任务信息。我们的 统一框架致力于将跨维度的信息和实验任务结合在一起 提供有意义的方式，以较低的维度解释所获得的信息。这些 发展将为数据科学和生物医学界提供新的工具 科学调查，包括用户友好的软件，以帮助医疗和公共卫生 基于生物行为多模式数据的决策。 目标。我们提出了三个具体的目标：1)(任务)开发一个特征分配框架 对通道内跨任务的高维生物行为数据进行建模；2) (纵向)扩展目标1的特征分配模型以考虑纵向 来自生物标志物的多个任务的数据的联合轨迹中的性能趋势 跨纵向访问的模式；3)(多模式)以跨整个数据建模 多模式生物标志物。每个目标中的建议都依赖于通过特征进行降维 在估计一组潜在的低维认知领域中的分配框架。 然后，根据儿童在代表不同因素的多个因素上的负荷量进行分组 认知域，有助于研究ASD的异质性。

PROJECT SUMMARY About 1 in 59 children are diagnosed with autism spectrum disorder (ASD), a neurodevelopmental disorder characterized by impairments in social interaction and communication. Our proposals in this grant are motivated by two studies on the two most promising biobehavioral biomarker modalities of ASD, electroencephalography (EEG) and eye- tracking (ET). Both studies collect data from serially administered EEG and ET tasks, over multiple longitudinal visits. In addition, multiple tasks within or across modalities tap into similar cognitive domains. Hence, even though joint analysis of these complex data structures across tasks, modalities (EEG and ET) and longitudinal visits would lead to the most efficient use of the available information, current analysis techniques are limited and are usually carried out on data from one task at a time, within a modality. Therefore, we propose a comprehensive set of statistical methods for the analysis of biobehavioral biomarker data in its entirety, borrowing information from multiple tasks, across modalities and over longitudinal visits. Our proposal relies on characterization of EEG and ET data as high-dimensional highly structured functional objects. Different from existing multimodal brain imaging literature, which fuses data for brain- region related inference, we combine a brain imaging modality (EEG) with a biobehavioral marker (ET), based on information on tasks that are related to common cognitive domains. Our unified framework strives to combine information across dimensions and experimental tasks to provide meaningful ways of interpreting the gained information in lower dimensions. These developments will provide the data science and biomedical community with novel instruments of scientific investigation, including user friendly software, to assist medical and public health decisions based on biobehavioral multimodal data. Aims. We propose three specific aims: 1) (Task) To develop a feature allocation framework for modeling the high-dimensional biobehavioral data across tasks within a modality; 2) (Longitudinal) To extend the feature allocation modelling of Aim 1 to account for longitudinal performance trends in the joint trajectories of data from multiple tasks of a biomarker within a modality across longitudinal visits; 3) (Multimodal) To model the data in its entirety across multimodal biomarkers. Proposals in each aim rely on dimension reduction through a feature allocation framework in estimating a set of underlying low-dimensional cognitive domains. Children are then clustered according to their loadings on multiple factors representing different cognitive domains, contributing to the study of heterogeneity in ASD.