Factors Affecting Prediction Accuracy of Complex Human Traits and Diseases

描述（由申请人提供）： 人类基因组计划的完成带来了基因组医学的希望——利用基因组信息来预防、诊断和治疗疾病。然而，尽管基因分型技术取得了巨大进步，但我们预测复杂人类特征和疾病的遗传易感性的能力仍然非常有限。我们自相矛盾地缺乏预测复杂人类特征的能力，部分原因可能在于全基因组关联研究中常用的统计方法所带来的局限性。我们相信，主要改编自动物育种领域的替代方法（WGP，全基因组预测）可以增强我们预测复杂人类特征和疾病的能力，从而为在个性化医疗中更深入地使用基因组信息铺平道路。 然而，WGP已成功应用的群体在选择历史、等位基因频率分布、连锁不平衡（LD）程度和近交等方面与人类群体存在很大差异。初步证据表明这些因素会影响 WGP 的预测性能。因此，需要利用人类数据对WGP进行综合评估，并且可能需要开发新的方法来应对复杂人类特征预测带来的挑战。 我们提出了一个框架来研究影响 WGP 解释和预测未观察 QTL 方差的能力的因素。使用该框架，结合模拟和真实数据分析，我们将利用人类数据对现有 WGP 进行首次综合评估，并量化数据关键特征、感兴趣的特征以及回归方法对现有 WGP 程序预测准确性的影响。我们将利用这些信息来开发新方法，以应对现有方法的局限性。 公共卫生相关性： 项目叙述尽管基因分型技术取得了巨大进步，但我们预测遗传风险的能力仍然非常有限。我们相信，主要源自动物育种的替代统计方法（WGP，全基因组预测）可能为提高我们预测重要健康结果的能力提供机会。然而，人类群体与动物和植物育种群体的不同之处可能会极大地影响 WGP 的预测性能。结合模拟和真实数据分析，我们将：（a）利用人类数据对现有 WGP 进行综合评估，（b）量化数据关键特征、感兴趣的特征以及回归方法对 WGP 预测准确性的影响，以及（c）开发新的回归程序，旨在应对现有回归程序中发现的局限性。

DESCRIPTION (provided by applicant): The completion of the human genome project brought the promise of genomic medicine-the use of genomic information for prevention, diagnosis and treatment of diseases. Yet, despite great progress in genotyping technologies, our ability to predict genetic predisposition to complex human traits and diseases remains very limited. Part of the explanation of our paradoxical lack of ability to predict complex human traits may reside in the limitations posed by the statistical methods commonly used in genome wide association studies. We believe that alternative methods, largely adapted from the field of animal breeding (WGP, whole-genome prediction), can enhance our ability to predict complex human traits and diseases, thus paving the way towards more intensive use of genomic information in personalized medicine. However, the populations to which WGP has been successfully applied differ greatly from human populations in aspects such as selection history, distribution of allele frequency, extent linkage disequilibrium (LD) and inbreeding. And preliminary evidence indicates that these factors can impact the predictive performance of WGP. Therefore, a comprehensive evaluation of WGP with human data is needed, and new methods may need to be developed to cope with the challenges posed by the prediction of complex human traits. We propose a framework to study the factors affecting the ability of WGP to account for and to predict variance at un-observed QTL. Using this framework, and a combination of simulation and real data analysis, we will produce the first comprehensive evaluation of existing WGP with human data and will quantify the effects of key features of the data, of the trait of interest, and of the regression method on the prediction accuracy of existing WGP procedures. We will use this information to develop new methods designed to confront the limitations of existing ones. PUBLIC HEALTH RELEVANCE: Project Narrative Despite great progress in genotyping technologies our ability to predict genetic risk remains very limited. We believe that alternative statistical methods (WGP, Whole Genome Prediction) largely adapted from animal breeding, may offer opportunities for advancing our ability to predict important health outcomes. However, human populations differ from animal and plant breeding populations in aspects that can greatly affect the predictive performance of WGP. Using a combination of simulations and real-data analysis we will: (a) produce a comprehensive evaluation of existing WGP with human data, (b) quantify the effects of key features of the data, of the trait of interest, and of the regression method on the prediction accuracy of WGP, and (c) develop new regression procedures designed to confront the limitations identified in existing ones.