Deep phenotyping in Electronic Health Records for Genomic Medicine

项目总结 该项目的总体目标是建立一个基因组医学学习系统来加速基因组 电子健康档案中的知识发现和应用。我们将融合深度特色 从EHR中提取的表型和关于基因-表型关联的不断发展的知识，以优化 变异解释的准确性和临床基因组/外显子组测序的成本效益，以及 通过构建动态的基因型-表型知识网络，加速因果基因的发现。 关于表型-基因关系的先验知识和患者的表型信息可以促进 在临床基因组学背景下从数千个基因变异中鉴定致病突变 测序；然而，如何最好地从符合以下条件的患者的EHR中提取表型信息 诊断为或评估为单基因疾病，标准化可计算的表示 表型，并将其用于基因组解释仍不清楚。此外，如何系统地比较 发现人类遗传学新知识的跨疾病表型在很大程度上仍是一个未开发的领域 带着巨大的希望。为了应对这些挑战，我们将开发和验证可伸缩和可移植的开源 自动准确提取特征表型的自然语言处理(NLP)方法 EHR叙述中的概念(例如，“J形鞍座”和“矮个子”)。然后，我们将开发一种 被称为EHR-Phenolyzer的表型驱动评分系统预测可能的候选遗传变异 与表型相关的患者的基因组测序和单基因突变的高概率 条件。在此基础上，我们将开发一个概率疾病诊断和知识发现系统，使用 丰富而深刻的EHR表型，并评估这些方法的基因组诊断和发现使用大- 标尺临床外显子组测序数据。最终，这些方法将支持高效、高效和可伸缩 基因组诊断，并促进基因组引导的精确医学在临床实践中的实施。

PROJECT SUMMARY The overarching goal of the project is to establish a genomic medicine learning system to accelerate genomic knowledge discovery and application in electronic health records (EHRs). We will integrate deep characteristic phenotypes extracted from EHRs and evolving knowledge of genotype-phenotype associations to optimize the accuracy of variant interpretation and the cost-effectiveness of clinical genome/exome sequencing, and to accelerate the discovery of causal genes by constructing a dynamic genotype-phenotype knowledge network. Prior knowledge on phenotype-gene relationships and phenotypic information about patients can facilitate the identification of disease-causing mutations from thousands of genetic variants in the context of clinical genomic sequencing; however, how best to abstract phenotype information from notes in the EHRs of patients who are diagnosed with or evaluated for monogenetic disorders, standardize the computable representation of phenotypes, and utilize it in genomic interpretation remains unclear. Additionally, how to systematically compare phenotypes across diseases to discover new knowledge in human genetics remains a largely untapped area with great promise. To address these challenges, we will develop and validate scalable and portable open-source natural language processing (NLP) methods for automated and accurate abstraction of characteristic phenotype concepts (e.g., “j-shaped sella turcica” and “short stature”) from EHR narratives. We will then develop a phenotype-driven scoring system called EHR-Phenolyzer to predict the likely candidate genetic variants associated with the phenotypes for patients with genomic sequencing and a high probability of a monogenic condition. On this basis, we will develop a probabilistic disease diagnosis and knowledge discovery system using rich and deep EHR phenotypes, and evaluate these methods for genomic diagnosis and discovery using large- scale clinical exome sequencing data. Ultimately, these methods will support efficient, effective, and scalable genomic diagnostics, and facilitate the implementation of genome-guided precision medicine in clinical practice.