Understanding the molecular mechanisms that contribute to neuropsychiatric symptoms in Alzheimer Disease

项目摘要 阿尔茨海默病（AD）是一种破坏性的神经退行性疾病，影响620万美国人，但目前 治疗不能有效预防或减缓认知能力下降1。神经精神症状（Neuropsychiatric symptoms，简称CNS） AD和相关痴呆的核心特征与日常功能的主要不良影响相关， 提高生活质量，加快机构化进程。母基金R 01 AG 067025的总体目标 是将单核转录组谱与来自每个供体的详细的SNP表型数据整合， 鉴定与疾病轨迹相关失调基因，鉴定具有不同基因的供体群 表达疾病特征，并提名用于新疗法靶向的基因和途径。 单核转录组谱的概要，包括来自约1，800个供体的约7.2M细胞核 由母基金R 01 AG 067025产生的资金是一个了不起的资源。然而挖掘这些转录组图谱 为了推进AD病因学的知识，需要分析工作流程， 以及其他新兴数据。多供体单细胞和细胞核转录组数据的现有工作流程 要么是1）为少数捐助者设计的，因此不能利用大规模和 此处使用的复杂研究设计，或2）改编自批量转录组分析，目前未扩展至 成百上千的捐赠者，几十种细胞类型和数百万个细胞。解决紧迫的生物问题的目标 关于AD生物学的假设需要开发设计和工程化的分析工作流程 考虑到多供体单细胞和细胞核转录组数据的挑战。 在本补充中，我们建议开发一个可扩展的，开源的分析工作流程，用于多供体单 细胞/细胞核转录组数据的动机，我们以前的工作线性混合模型2，3。我们先前已经 应用线性混合模型分析批量转录组谱，并开发了开源 variancePartition包，用于执行差异表达测试，考虑技术批次效应， 表征表达变异的多种生物和技术来源。虽然目前的软件 促进了我们小组和其他许多人对大量转录组和表观基因组图谱的分析， 多供体单核数据目前受到方差分区代码库的特别设计的限制。 为了解决这些限制，我们在这里提出（目标1）将此分析工作流程扩展到新兴数据集 使用软件工程、代码重构和跨多个计算环境的经验测试中的最佳实践 （目标2）使（a）计算生物学家能够更广泛地使用， 说明软件在公共数据集上的应用，并由（B）开源开发人员通过改进代码 设计和文档。总体而言，重新定义方差分区的分析工作流程将使 强大的线性混合模型方法，可扩展到多供体单细胞和细胞核转录组数据集， 以解决有关AD病因学的问题，并作为更广泛社区的开源工具。

PROJECT SUMMARY Alzheimer's disease (AD) is a devastating neurodegenerative disease that affects 6.2M Americans, yet current therapies are not effective at preventing or slowing the cognitive decline1. Neuropsychiatric symptoms (NPS) are core features of AD and related dementias that are associated with major adverse effects on daily function and quality of life, and accelerate time to institutionalization. The overarching goal of the parent grant R01AG067025 is to integrate single nucleus transcriptome profiles with detailed NPS phenotype data from each donor and identify dysregulated genes associated with disease trajectory, identify clusters of donors with different gene expression disease signatures, and nominate genes and pathways for targeting with novel therapeutics. The compendium of single nucleus transcriptome profiles comprising ~7.2M nuclei from ~1,800 total donors generated by the parent grant R01AG067025 is a remarkable resource. Yet mining these transcriptome profiles to advance knowledge of AD etiology requires analytical workflows that scale to the unprecedented size of these and other emerging data. Existing workflows for multi-donor single cell and nucleus transcriptome data have either been 1) designed for a small number of donors and so cannot take advantage of the large-scale and complex study design used here, or 2) adapted from bulk transcriptome analyses and do not currently scale to hundreds of donors, dozens of cell types and millions of cells. The objective of addressing pressing biological hypotheses about AD biology necessitates the development of analytical workflows designed and engineered with the challenges of multi-donor single cell and nucleus transcriptome data in mind. In this Supplement, we propose developing a scalable, open source analytical workflow for multi-donor single cell/nucleus transcriptome data motivated by our previous work on linear mixed models2,3. We have previously applied linear mixed models to analyze bulk transcriptome profiles, and developed the open source variancePartition package to perform differential expression testing, account for technical batch effects and characterize the multiple biological and technical sources of expression variation. While the current software has facilitated analysis of bulk transcriptomic and epigenomic profiles by our group and many others, applying it to the multi-donor single nucleus data is currently limited by the ad hoc design of the variancePartition codebase. To address these limitations, here we propose (Aim 1) Scaling this analytical workflow to emerging datasets using best practices in software engineering, code refactoring, and empirical testing across multiple computing environments; and (Aim 2) Enabling broader use by (a) computational biologists by developing vignettes to illustrate applications of the software on public datasets, and by (b) open source developers by improving code design and documentation. Overall, reconceiving the analytical workflow of variancePartition will enable the powerful linear mixed model approach to scale to multi-donor single cell and nucleus transcriptome datasets in order to address questions about the etiology of AD and serve as an open source tool for the broader community.