Multi-omic Risk Prediction of Chronic Obstructive Pulmonary Disease in European- and African-Ancestry Populations

项目摘要/摘要 慢性阻塞性肺疾病(COPD)是全世界呼吸系统死亡的主要原因15。 在病程早期识别高易感个体并了解致病原因 在不可逆转的肺功能丧失之前，机制是最重要的16，17。 40%的COPD易感性18-20。全基因组关联研究(GWAS)发现了多种变异 与COPD21-23有关。单独的变异很难预测风险，但总体来说，遗传变异可以。 占风险的很大一部分。汇集了数以百万计的GWAS变种，我创建了一个多基因风险评分(PRS) 对于COPD，这可以识别出COPD的高危个体，尽管在非 欧洲人24.由于在临床实践中不容易确定遗传祖先，因此需要多个祖先的PRS。 此外，反映遗传和环境影响的基因表达提供了病理生物学信息。 对于COPD易感性和异质性。增加预测价值的COPD转录风险评分(TRS) 以上临床危险因素25尚待开发。使用OMICS数据进行风险分层的吸引力在于 这些数据可以帮助我们了解为什么某些COPD亚组的进展风险更高。基因 监管网络26已经被用来揭示COPD异质性的机制，这些机制没有被 传统的基于基因的方法。因此，我们假设多基因和转录风险分数将 在识别COPD高危人群和相关表型方面显著改善临床因素， 并可用于识别治疗干预的途径。我们将使用4225人培训多血统的PRS 来自英国生物库的非洲血统个人和现有的来自Charge的8429名非洲裔美国人的分析， COPD(COPD基因：n=10,198)遗传流行病学和肺组织研究中的检测 财团(LTRC：N=1,078)。我们将使用全血创建多血统转录风险评分(TRS) 训练中的RNA测序(RNA-seq)数据(n=3,394)，并评估测试样本中的预测性能 COPD基因(n=1,131)。我们将使用连接图(Cmap)8，27来识别药物再用途候选 在TRS的记录上。我们将利用来自LTRC的肺RNA-seq数据来创建肺TRS，并在COPDgene中进行测试 血样。我们将沿着现有的PRS和肺TRS的轴向对COPDgene参与者进行分类(例如： “高”PR，“低”TRS)，我们预计这将识别COPD相关表型的高危人群，并 进步。为了了解为什么某些人患COPD表型的风险很高，我们将利用基因 调控网络，以识别不同的PRS/TRS分类之间的途径，并使用基因 监管网络数据库(GRAND)9，以确定药物再用途候选的优先顺序。这些目标将产生 未来研究的数据，重点是验证COPD-Omics风险分数和现实世界中的候选药物 队列1，使用机器学习来预测候选药物的网络效应。拟议的研究 职业发展计划将训练我使用机器学习进行多组集成和风险预测。

PROJECT SUMMARY/ABSTRACT Chronic obstructive pulmonary disease (COPD) is a leading cause of respiratory mortality worldwide15. Identifying highly susceptible individuals early in their disease course and understanding pathogenic mechanisms, before irreversible loss of lung function, is of utmost importance16,17. Genetics account for about 40% of COPD susceptibility18–20. Genome-wide association studies (GWASs) have identified multiple variants associated with COPD21–23. Individual variants are poor for risk prediction, but in aggregate genetic variants can account for a substantial portion of risk. Pooling millions of GWAS variants, I created a polygenic risk score (PRS) for COPD that can identify individuals at high risk for COPD, though performance was less optimal in non- Europeans24. Multi-ancestry PRSs are needed as genetic ancestry is not readily determined in clinical practice. Further, gene expression, reflecting genetic and environmental influences, provides pathobiologic information for COPD susceptibility and heterogeneity. A transcriptional risk score (TRS) for COPD that adds predictive value above clinical risk factors25 has yet to be developed. The appeal of using -Omics data for risk stratification is that these data can lend insight into why certain COPD subgroups are at elevated risk of progression. Gene regulatory networks26 have been used to uncover mechanisms of COPD heterogeneity that were not found by traditional gene-based approaches. Therefore, we hypothesize that polygenic and transcriptional risk scores will substantially improve upon clinical factors in identifying those at higher risk for COPD and related phenotypes, and can be used to identify pathways for therapeutic intervention. We will train multi-ancestry PRSs using 4,225 African ancestry individuals from UK Biobank and existing analyses of 8,429 African-Americans from CHARGE, and test in the Genetic Epidemiology of COPD (COPDGene: n=10,198) study and Lung Tissue Research Consortium (LTRC: n=1,078). We will create a multi-ancestry transcriptional risk score (TRS) using whole blood RNA-sequencing (RNA-seq) data in training (n=3,394) and evaluate predictive performance in testing samples (n=1,131) of COPDGene. We will use Connectivity Map (CMap)8,27 to identify drug repurposing candidates based on TRS transcripts. We will leverage lung RNA-seq data from LTRC to create a lung TRS, and test in COPDGene blood samples. We will classify COPDGene participants along the axes of the existing PRS and lung TRS (e.g. “High” PRS, “Low” TRS), which we expect will identify those at high risk for COPD-related phenotypes and progression. To understand why certain individuals are at high risk for COPD phenotypes, we will utilize gene regulatory networks to identify pathways differing between PRS/TRS classifications, and use the Gene RegulAtory Network Database (GRAND)9 to prioritize drug repurposing candidates. These aims will generate data for future studies, which will focus on validating COPD -Omics risk scores and drug candidates in real-world cohorts1, and using machine learning to predict the network effects of drug candidates. The proposed research and career development plan will train me to use machine learning for multi-omic integration and risk prediction.