Genetics of deep-learning-derived neuroimaging endophenotypes for Alzheimer's Disease

阿尔茨海默病(AD)的特征是进行性的认知和 记忆功能正常，是老年人最常见的痴呆症形式。它影响560万人 65岁以上的美国人对患者提出了巨大且日益增长的要求， 护理者和医疗资源，使这种情况成为最重要的公共疾病之一 我们这个时代的健康问题。尽管进行了广泛的研究，但我们对生物学和 阿尔茨海默病的病理生理机制仍然有限，阻碍了治疗和治疗的进展 预防策略。阿尔茨海默病的遗传学研究已成功识别出40个新的基因座，但这些 只解释总体疾病风险的一小部分，建议有更多机会 发现。先进的神经成像是当前AD临床和研究的重要组成部分 研究，一般集中在相对较少的成像表型由神经- 放射科医生。然而，人们对利用中的高内容信息的兴趣越来越大 大规模、高维多模式神经成像数据用于识别新的AD生物标记物。 深度学习方法是机器学习研究的一个新兴领域，它使用原始图像 以获得成像内容的最佳矢量表示，其可用作信息 AD内表型。要克服传统上归因于DL、Whole的低可解释性 基因组序列数据提供了一个机会，以确定新的基因潜在的DL- 衍生的成像内表型及其与AD和AD相关性状的相关性 大量的队列样本。拟议的项目将利用现有的神经成像和基因 数据资源来自英国生物库、阿尔茨海默病测序项目(ADSP)、 阿尔茨海默病神经成像倡议(ADNI)和心脏与老龄化队列 基因组流行病学研究(CHARE)联盟，并将由 由多学科调查人员组成的团队。我们将从神经成像中得出AD的内表型 英国生物库中使用深度学习(DL)的数据。我们将识别与之相关的新的基因座 具有DL衍生的成像内表型，并优化这些 使用英国生物库遗传数据的AD相关表型的内表型。我们将利用 资源和与AD联盟的合作以及DL派生的神经成像的力量 确定阿尔茨海默病和AD相关特征的新基因的内表型。另外，我们 将开发基于DL的神经成像协调和归因方法并分发 向研究界提供实施软件。我们希望发现新的相关基因 这可能有助于了解AD的分子基础和潜在的新的治疗方法。

Alzheimer's disease (AD) is characterized by the progressive impairment of cognitive and memory functions and is the most common form of dementia in the elderly. It affects 5.6 million Americans over the age of 65 and exacts tremendous and increasing demands on patients, caregivers, and healthcare resources, making this condition among the most significant public health problems of our time. Despite extensive studies, our understanding of the biology and pathophysiology of AD is still limited, hindering advances in the development of therapeutic and preventive strategies. Genetic studies of AD have successfully identified 40 novel loci but these explain only a fraction of the overall disease risk, suggesting opportunities for additional discoveries. Advanced neuroimaging is an essential part of current AD clinical and research investigations, which generally focus on relatively few imaging phenotypes developed by neuro- radiologists. However, there is a growing interest in exploiting the high-content information in large-scale, high dimensional multimodal neuroimaging data to identify novel AD biomarkers. Deep learning (DL) methods, an emerging area of machine learning research, uses raw images to derive optimal vector representations of imaging contents, which can be used as informative AD endophenotypes. To overcome the low interpretability traditionally attributed to DL, whole genome sequence data provide an opportunity to identify novel genes underlying the DL- derived imaging endophenotypes and test their association with AD and AD-related traits in large cohort samples. The proposed project will leverage existing neuroimaging and genetic data resources from the UK Biobank, the Alzheimer's Disease Sequencing Project (ADSP), the Alzheimer's Disease Neuroimaging Initiative (ADNI), and the Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE) consortium, and will be conducted by a multidisciplinary team of investigators. We will derive AD endophenotypes from neuroimaging data in the UK Biobank using deep learning (DL). We will identify novel genetic loci associated with DL-derived imaging endophenotypes and optimize the co-heritability of these endophenotypes with AD-related phenotypes using UK Biobank genetic data. We will leverage resources and collaborations with AD Consortia and the power of DL-derived neuroimaging endophenotypes to identify novel genes for Alzheimer's Disease and AD-related traits. Also, we will develop DL-based neuroimaging harmonization and imputation methods and distribute implementation software to the research community. We expect to discover new genes relevant to AD which may leads to understanding of molecular basis of AD and potential new treatment.