Glaucoma Risk Prediction Using Machine Learning Integration of Image-Based Phenotypes and Genetic Associations

项目总结/摘要 该提案描述了一个为期5年的培训计划，以发展学术生涯，重点是提高 通过基因组和表型风险的组合进行青光眼风险预测。我会用有监督的半- 定义新颖结构和纵向图像的监督和非监督机器学习方法 基于POAG的内表型与疾病亚型和进展一致。这些内表型将 用于发现新的疾病相关基因座。通过包括纵向数据，我们的目标是确定 进行性疾病的遗传标记我们将使用已知的POAG风险变异和新的遗传变异 在这些分析中鉴定的多基因风险评分（PRS）用于创建POAG的几个候选全基因组多基因风险评分（PRS）。 将测试添加和不添加人口统计学和图像特征的每个候选PRS的效用， 预测青光眼风险是独立的NEIGHBORHOOD和LIFE队列。我们假设基于PRS的 与我们的内表型相关的遗传变异将提高POAG病例预测能力 与基于横截面全基因组关联研究的PRS相比。拟议的研究包括 有可能深入了解疾病的发病机制，并提高基因检测的预测能力 我完全有能力进行这项研究，并接受这里提出的培训。我有一种强烈 具有定量科学背景，工程学位，统计培训和既定的跟踪记录 大型数据库研究。此外，我还提出了一个详细的职业发展计划， 1）学习基于机器学习的自动化方法的基础知识，应用程序和限制 眼底图像分析和2）了解计算生物学和统计方法，以处理大型 基因组学数据集。我的培训计划包括在HSPH的定量方法MPH与浓度 在计算生物学和统计学习中。此外，我还得到了一个多学科团队的支持， 致力于我的学术成长和发展成为一个独立的临床科学家的导师。我 我将与青光眼遗传学专家Wiggs和塞格雷博士以及统计学和机器学领域的领导者合作， 学习，埃尔兹博士和卡尔帕蒂-克莱默博士。我将有充分的机会获得广泛的资源， 医疗保健和哈佛系统为这项工作和我的职业发展。 本文概述的研究将提高我们对青光眼发病机制的理解， 青光眼筛查的多模式精准医学方法开发基金会， 诊断.这项研究是尖端的，为我作为一个独立的NIH资助的职业生涯做好了准备。 研究者的目的是使用纵向多模态临床，成像，测试和多组学数据， 少数民族青光眼患者1）了解视力丧失的途径，2）开发精确的医学方法 在术前识别功能性视力丧失和进展的高风险患者，3）使这些 技术的临床现实，以减少不必要的失明的负担。

PROJECT SUMMARY/ ABSTRACT This proposal describes a 5-year training program to develop an academic career focused on improving glaucoma risk prediction through a combination of genomic and phenotypic risk. I will use supervised, semi- supervised and unsupervised machine learning methods to define novel structural and longitudinal image based endophenotypes for POAG aligned with disease subtype and progression. These endophenotypes will be used to discover new disease associated genomic loci. By including longitudinal data, we aim to identify genetic markers for progressive disease. We will use known POAG risk variants and novel genetic variants identified in these analyses to create several candidate genome wide polygenic risk scores (PRS) for POAG. Each candidate PRS with and without addition of demographic and image features will be tested for its utility to predict glaucoma risk is independent NEIGHBORHOOD and LIFE cohorts. We hypothesize that a PRS based on genetic variants associated with our endophenotypes will have improved POAG case predictive power compared to PRS based on cross-sectional genome wide association studies. The proposed studies have the potential to provide insight into disease pathogenesis and improve predictive power of genetic testing I am well positioned to conduct this research and undertake the training proposed here. I have a strong quantitative science background with a degree in engineering, statistical training and established track records of large database research. Additionally, I have proposed a detailed career development plan that will allow me to 1) learn the fundamentals, applications and limitations of machine learning based approaches for automated fundus image analysis and 2) understand computational biology and statistical approaches to handle large genomics datasets. My training plan includes an MPH in quantitative methods at the HSPH with concentration in computational biology and statistical learning. Additionally, I am supported by a multidisciplinary team of committed mentors dedicated to my academic growth and progression into an independent clinician scientist. I will work with glaucoma genetics experts, Drs Wiggs and Segre, and leaders in statistical and machine learning, Drs Elze and Kalpathy-Cramer. I will have full access to the extensive resources at MEE, Partners Healthcare and the Harvard system for this work and my career development. The research outlined here will improve our understanding of glaucoma pathogenesis and lay the foundation for development of multimodal precision medicine approaches for glaucoma screening and diagnosis. This research is cutting edge and prepares me well for my career as an independent NIH funded investigator with the aim to use longitudinal multi-modal clinical, imaging, testing and multi-omics data in multi- ethnic glaucoma patients to 1) understand pathways of vision loss, 2) develop precision medicine approaches to pre-symptomatically identify patients at high risk of functional vision loss and progression and 3) make these technologies a clinical reality in order to reduce the burden of unnecessary blindness.