Multivariate Pattern Analysis Methods for Neuroimaging Genetics Studies

描述（由申请人提供）：常见的精神障碍，如阿尔茨海默病和精神分裂症，在很大程度上是遗传的，具有复杂的遗传基础。大规模全基因组关联研究对比了来自患者和对照组的DNA序列数据，最近发现了这些疾病的新的遗传风险变异。然而，基因型增加风险的过程尚未得到充分表征。神经影像学提供了比临床诊断更丰富的潜在疾病过程的图像。因此，神经影像学和遗传学数据的联合分析有望促进我们对这些过程的理解。然而，今天，神经成像遗传学研究面临着阻碍进展的重要挑战：小样本量、适度的效应量和数据的极端维度限制了统计能力，从而限制了我们探索基因、神经解剖学和临床衰退之间复杂而微妙的联系的能力。目前，在神经成像遗传学中流行的方法是集中分析少量感兴趣的解剖区域和/或候选基因，而经常忽略大部分数据。该项目的核心目标是开发计算工具，充分利用数据集的丰富性，促进探索基因型、神经影像学测量和临床表型之间的多方面关联。拟议的项目将使用先进的多元模式分析方法，如支持向量机来计算反映病理的基于图像和遗传评分。我们将基于它们与经典疾病生物标志物的关联来验证这些工具。最后，我们将开发一个模型，使用成像和基因型数据来预测未来的临床结果。我们希望这些工具能够在与复杂精神障碍相关的三个方向上取得进展，例如迟发性阿尔茨海默病（AD）：(1)确认和表征风险基因，(2)识别健康个体的疾病特异性解剖改变，以及(3)早期诊断和预后。该项目将(1)使用三个已经收集的大规模数据集将开发的工具应用于AD，(2)建立在我们一直在开发的尖端图像处理算法上，(3)允许候选人接受神经解剖学，精神障碍和遗传学的进一步培训，为他未来作为独立研究人员的职业生涯奠定基础。

DESCRIPTION (provided by applicant): Common mental disorders such as Alzheimer's disease and schizophrenia are largely heritable with complex genetic underpinnings. Large-scale genome-wide association studies that contrast DNA sequence data from patients and controls have recently identified novel genetic risk variants for these disorders. Nevertheless, the processes through which genotype increases risk are yet to be fully characterized. Neuroimaging offers a richer picture of the underlying disease processes than a clinical diagnosis. Thus the joint analysis of neuroimaging and genetics data promises to advance our understanding of these processes. Today, neuroimaging genetics studies however face important challenges that obstruct progress: small sample sizes, modest effect sizes, and the extreme dimensionality of the data limit statistical power and thus our ability to explore the complex and subtle associations between genes, neuroanatomy and clinical decline. Currently, the prevalent approach in neuroimaging genetics is to concentrate the analysis on a small number of anatomic regions of interest and/or candidate genes and often ignore a large portion of the data. The core goal of the proposed project is to develop computational tools that will take full advantage of the richness in the datasets and facilitate the exploration of the multifaceted associations between genotype, neuroimaging measurements and clinical phenotype. The proposed project will use advanced multivariate pattern analysis methods such as support vector machines to compute image-based and genetic scores that reflect pathology. We will validate the tools based on their association with classical biomarkers of disease. Finally, we will develop a model that uses both imaging and genotype data to predict future clinical outcome. We expect these tools will enable progress along three directions relevant to complex mental disorders, e.g. late-onset Alzheimer's disease (AD): (1) confirming and characterizing risk genes, (2) identifying disease-specific anatomical alterations in healthy individuals, and (3) early diagnosis and prognosis. The project will (1) use three already-collected large-scale datasets to apply the developed tools to AD, (2) build on cutting-edge image processing algorithms that we have been developing, and (3) allow the candidate to receive further training in neuroanatomy, mental disorders and genetics, forming the foundation for his future career as an independent researcher. PUBLIC HEALTH RELEVANCE: Project Narrative/Relevance We will develop computational tools for analyzing complex associations between images, genotype and clinical phenotype. The tools will be user-friendly and freely available, and will potentially facilitate accurate early diagnosis and prognosis of mental disorders such as Alzheimer's.