Bioinformatics Strategies for Multidimensional Brain Imaging Genetics

描述(由申请人提供)： 今天的多模式成像系统产生了海量的高维数据集，当这些数据集与高通量的基因分型数据(如单核苷酸多态(SNPs))相结合时，提供了令人兴奋的机会来增强我们对人类疾病的表型特征和遗传结构的理解。然而，这些数据集前所未有的规模和复杂性造成了严重的计算瓶颈，需要新的概念和使能工具。为了应对这些挑战，以阿尔茨海默病(AD)研究为试验台，该项目将为多维脑成像遗传学开发和验证新的生物信息学策略。目标1是开发一种新的双多变量分析策略，S3K-CCA，用于研究成像遗传关联。现有的成像遗传学方法通常被设计来发现单SNP-单QT、单SNP-多QT或多SNP-单QT的关联，在揭示相互关联的遗传标记和相关的脑表型之间的复杂关系方面能力有限。为了克服这一局限性，S3K-CCA被设计成一个稀疏的双多变量学习模型，它同时使用多个响应变量和多个预测因子来分析大规模的多模式神经基因组数据。目标2是开发HD-BIG，这是一个可视化和系统生物学框架，用于综合分析高维大脑成像遗传学数据。机器学习策略无缝地整合有价值的领域知识以产生生物学上有意义的结果，仍然是成像遗传学中一个未被充分探索的领域。为此，我们将开发一个用户友好的热图界面来可视化高维结果， 调整学习参数和策略，与现有生物信息学资源和工具互动，促进视觉探索性和系统生物学分析。一种新的成像遗传富集法(IGEA)将被开发出来，以识别相关的遗传途径和相关的脑回路，并揭示它们之间的复杂关系。目的3是使用模拟和真实的成像遗传学数据来评估所提出的S3K-CCA和IGEA方法以及HD-BIG框架。该项目有望产生新的生物信息学算法和工具，用于大规模异质成像遗传学数据的综合联合分析。这些强大的方法的可用性对于许多成像遗传学计划的成功至关重要。此外，它们还可以帮助在生物医学研究的其他领域实现新的计算应用，在这些领域，对大规模多模式数据的系统和综合分析至关重要。以阿尔茨海默病为例，建议的方法将展示加强对复杂疾病的机械性理解的潜力，这可以通过促进诊断和治疗进展而有利于公共卫生结果。

DESCRIPTION (provided by applicant): Today's generation of multi-modal imaging systems produces massive high dimensional data sets, which when coupled with high throughput genotyping data such as single nucleotide polymorphisms (SNPs), provide exciting opportunities to enhance our understanding of phenotypic characteristics and the genetic architecture of human diseases. However, the unprecedented scale and complexity of these data sets have presented critical computational bottlenecks requiring new concepts and enabling tools. To address these challenges, using the study of Alzheimer's disease (AD) as a test bed, this project will develop and validate novel bioinformatics strategies for multidimensional brain imaging genetics. Aim 1 is to develop a novel bi- multivariate analysis strategy, S3K-CCA, for studying imaging genetic associations. Existing imaging genetics methods are typically designed to discover single-SNP-single-QT, single-SNP-multi-QT or multi-SNP-single- QT associations, and have limited power in revealing complex relationships between interlinked genetic markers and correlated brain phenotypes. To overcome this limitation, S3K-CCA is designed to be a sparse bi- multivariate learning model that simultaneously uses multiple response variables with multiple predictors for analyzing large-scale multi-modal neurogenomic data. Aim 2 is to develop HD-BIG, a visualization and systems biology framework for integrative analysis of High-Dimensional Brain Imaging Genetics data. Machine learning strategies to seamlessly incorporate valuable domain knowledge to produce biologically meaningful results is still an under-explored area in imaging genetics. In this aim, we will develop a user-friendly heat map interface to visualize high-dimensional results, adjust learning parameters and strategies, interact with existing bioinformatics resources and tools, and facilitate visual exploratory and systems biology analysis. A novel imaging genetic enrichment analysis (IGEA) method will be developed to identify relevant genetic pathways and associated brain circuits, and to reveal complex relationships among them. Aim 3 is to evaluate the proposed S3K-CCA and IGEA methods and the HD-BIG framework using both simulated and real imaging genetics data. This project is expected to produce novel bioinformatics algorithms and tools for comprehensive joint analysis of large scale heterogeneous imaging genetics data. The availability of these powerful methods is critical to the success of many imaging genetics initiatives. In addition, they can also help enable new computational applications in other areas of biomedical research where systematic and integrative analysis of large-scale multi-modal data is critical. Using AD as an exemplar, the proposed methods will demonstrate the potential for enhancing mechanistic understanding of complex disorders, which can benefit public health outcomes by facilitating diagnostic and therapeutic progress.