Learning the Regulatory Code of Alzheimer's Disease Genomes

项目摘要 选择性剪接是一个关键的细胞过程，其失调已被广泛牵连 人类遗传疾病的研究。诺尔斯博士以前开发了LeafCutter，一种灵活的，可扩展的， 这是一个无注释的工具，可以从短读RNA-seq数据中量化RNA剪接的局部模式。 虽然LeafCutter已经被广泛采用，我们也积极维护它， 虽然在github上解决了问题，但它仍然是“早期阶段”的软件。我们提出软件 工程改进：1）使用具有标准构造的conda的适当包装， 安装、测试和日志记录流程，2）使用Docker的容器化，3）标准化 输入/输出数据格式/接口，以及4）重构以使用标准工作流程 语言这些改进将允许我们通过存储库分发LeafCutter 包括PyPI、BioConda、DockStore和Galaxy。最后是对文档的改进， 测试和版本管理将从开放源代码为LeafCutter做出贡献 社区更可行，更容易融入。 建议工作的家长奖是U 01 AG 068880 -01“学习监管代码， 阿尔茨海默病基因组”，我们正在开发最先进的深度学习（DL） 和机器学习（ML）模型，以更好地了解AD的遗传基础。这个奖项 广泛使用LeafCutter。在目标1中，我们正在构建前和后的DL模型， 转录调控：对于后者，LeafCutter为我们的神经网络提供了训练数据。 AD相关细胞类型中RNA剪接序列决定簇的网络模型， states.在目标2中，我们将AD相关的结构变异与功能变异联系起来，包括 RNA剪接变异。在目标3中，我们将构建数千个反式表达QTL网络， 死后大脑样本：对LeafCutter生态系统的改进 在这里，我们将能够直接扩展到反式剪接QTL网络。虽然我们和 我们的合作者本身就是LeafCutter的重度用户，我们将继续确保我们 提供更广泛的基因组学社区的需求和用例。

Project Summary Alternative splicing is a key cellular process whose dysregulation has been implicated broadly across human genetic disease. Dr. Knowles previously developed LeafCutter, a flexible, scalable, annotation-free tool to quantify local patterns of RNA splicing from short-read RNA-seq data. While LeafCutter has been quite widely adopted and we have actively maintained it and addressed issues on github, it remains “early stage” software. We propose software engineering improvements: 1) appropriate packaging using conda with standard build, installation, testing and logging processes, 2) containerization using Docker, 3) standardization of input/output data formats/interfaces, and 4) refactoring to use a standard workflow language. These improvements will allow us to distribute LeafCutter through repositories including PyPI, BioConda, DockStore, and Galaxy. Finally improvements to documentation, testing and version management will make contributions to LeafCutter from the open source community more feasible and easier to integrate. The parent award for the proposed work is U01 AG068880-01 “Learning the Regulatory Code of Alzheimer's Disease Genomes”, where we are developing state-of-the-art deep learning (DL) and machine learning (ML) models to better understand the genetic basis of AD. This award makes extensive use of LeafCutter. In Aim 1 we are building DL models of pre- and post- transcriptional regulation: for the latter LeafCutter provides training data for our neural network model of the sequence determinants of RNA splicing in AD-relevant cell types and states. In Aim 2 we connect AD-associated structural variation to functional variation, including RNA splicing variation. In Aim 3, we will build trans-expression QTL networks across thousands of post-mortem brain samples: with the improvements to the LeafCutter ecosystem proposed here we will be able to straightforwardly extend to trans splicing QTL networks. While we and our collaborators are ourselves heavy users of LeafCutter, we will continue to ensure we provide for the needs and use-cases of the broader genomics community.