Bioinformatics Strategies for Genome Wide Association Studies

精确医学的承诺是编辑患者的DNA和/或实施针对病因的治疗 使用量身定做的设计来预防或逆转疾病过程的分子。所有这一切都发生在 并且需要对患者的生物学有准确的了解。与之形成鲜明对比的是，许多 我们拥有的关于基因组风险因素的知识来自于人类相关性的统计测量 人口。我们研究的群体和我们所研究的个体之间的概念和实践上的脱节 想要治疗是一个困惑的主要来源，在一个由基因组驱动的时代如何前进 技术这项提议的主要目标是开发新的信息学方法和软件来 通过将群体和个体基因组现象联系起来，促进精准医学。我们在这里提出一个 虚拟基因组医学(VGMed)工作台，临床医生可以在其中进行关于 使用从人群水平研究得出的疾病风险模型对个别患者进行治疗。这将是 通过首先开发一种新的基因组学指导的自动机器学习(GAML)算法来实现 从临床医生可访问的真实数据中导出风险模型(目标1)。然后我们将开发一部小说 一种模拟方法，能够生成保存遗传效应分布的人工数据 在真实数据中观察，同时保持其他特征，如基因频率(AIM 2)。这 将生成开放数据，允许任何人对来自人群的患者进行虚拟干预- 水平风险分布。工作台将允许编辑单个基因类型并模拟给药 通过在模拟模型中编辑机器学习参数来对药物进行分类(AIM 3)。风险和风险的变化 将实时跟踪特定患者的疾病状态。最后，我们在工作台中提供了一个特性 这将允许临床医生产生关于单个基因变异的特定假设，然后这些假设可以 使用包括PubMed和ClinVar等数据库的集成知识来源进行验证，从而提供 用户即时反馈(目标4)。所有方法和软件都将以开源形式提供(AIM 5)。

The promise of precision medicine is to edit a patient’s DNA and/or administer therapeutics targeting etiologic molecules that prevent or reverse the disease process using a tailored design. All of this happens at the level of the individual and requires precision knowledge of that patient’s biology. In stark contrast, much of the knowledge we possess about genomic risk factors comes from statistical measures of association from human populations. The conceptual and practical disconnect between the populations we study and the individuals we want to treat is a major source of confusion about how to move forward in an era driven by genome technology. The primary goal of this proposal is to develop novel informatics methodology and software to facilitate precision medicine by connecting population and individual genomic phenomena. We propose here a Virtual Genomic Medicine (VGMed) workbench where clinicians can carry out thought experiments about the treatment of individual patients using models of disease risk derived from population-level studies. This will be accomplished by first developing a novel Genomics-guided Automated Machine Learning (GAML) algorithm for deriving risk models from real data that is accessible to clinicians (AIM 1). We will then develop a novel simulation approach that is able to generate artificial data that preserves the distribution of genetic effects observed in the real data while maintaining other characteristics such as genotype frequencies (AIM 2). This will generate open data allowing anyone to perform virtual interventions on patients derived from a population- level risk distribution. The workbench will allow editing of individual genotypes and simulate the administration of drugs by editing machine learning parameters in the simulation model (AIM 3). The change in risk and disease status for the specific patient will be tracked in real time. Finally, we provide a feature in the workbench that will allow the clinician to generate specific hypotheses about individual genetic variants that can then be validated using integrated knowledge sources that include databases such as PubMed and ClinVar thus giving the user immediate feedback (AIM 4). All methods and software will be provided as open-source (AIM 5).