Genomic prediction tools developed using phenotypes from disease progression models

 描述(由申请人提供)： 我之前的培训和长期的职业目标使这个药物基因组学和个性化医学翻译学者奖(K23)成为帮助我成为一名独立研究员的理想机会。在过去的三年里，我在芝加哥大学完成了血液学/肿瘤学、临床药理学和药物基因组学的联合奖学金。我的研究员研究向我介绍了药物计量学领域，这是一门量化药物、疾病和试验特征的科学。具体地说，我对使用药物计量学来指导癌症治疗的开发和优化使用产生了兴趣。在过去的两年里，作为一名Paul Calabresi K12学者，我进一步促进了这种兴趣，在此期间，我从这个国家仅有的几个药物计量学项目之一获得了药物计量学硕士学位 马里兰大学巴尔的摩分校。与此同时，我获得了丰富的写作经验和 进行主要药物基因组成分的临床试验。为了推进我作为一名独立研究员的职业生涯，我现在询问药物计量学是否可以提高将基因组数据转化为临床决策的能力。 已经有许多药物基因组学发现描述了种系遗传变异和药物毒性或疗效之间的联系。关于药效的进一步发现可以通过两种方式得到加强：第一，通过确定更好的药效表型；第二，通过在预测工具中利用大量常见的变种，而不是依赖于少数变种。在我的合作研究中，我们开发了一个肾细胞癌(RCC)的疾病进展模型，可以估计药物在人群和个体患者中的治疗效果。这种模型估计的治疗效果是一个有趣的潜在表型，因为它考虑了关于肿瘤大小的全套纵向数据，而不是更传统的表型，如客观反应(肿瘤大小减少≥30%)、无进展存活率和总存活率。合作者已经证明，半自动测量肿瘤体积，而不是横断面图像上的最长直径，可能会通过提供更准确的肿瘤负担评估来进一步增强这些表型。最后，我在芝加哥大学的同事开发了一种方法(称为OmicKriging)，可以使用在全基因组询问(以及潜在的其他基因组数据)中发现的所有常见变异来预测表型。 CORZZ试验是有史以来规模最大的转移性肾癌随机第三阶段试验，共有1110名患者在北美、欧洲和亚洲随机接受帕佐帕尼和舒尼替尼的治疗。其中约三分之二的患者提供了胚系DNA用于全基因组基因分型，这一工作已经完成。作为与赞助这项试验的葛兰素史克制药公司(GSK)持续合作的一部分，我们可以访问这些患者的临床数据、计算机断层扫描图像和全基因组基因分型数据。这些数据提供了一个独特的机会，使用两种新的治疗方法和一种新的表型(纵向肿瘤体积)来修订我们以前的肾癌疾病进展模型，并探索如何使用常见的变异来预测模型估计的治疗效果和传统表型，如客观反应、PFS和OS。这个假设是，通过疾病进展模型估计的表型将导致比传统表型更好的基因组预测工具。这些工具可以预测哪些患者或多或少可能从转移性肾癌的酪氨酸激酶抑制剂治疗中受益。此外，这些工具可以通过添加其他数据元素(如肿瘤基因)来改进，并可以作为在其他癌症和其他复杂疾病中使用基于模型的表型的类似工具的蓝图。 在研究计划中，我描述了四个步骤(目标)，从逻辑上将我们从原始数据带到经过验证的基于模型和传统表型的基因组预测工具。第一步是通过测量肿瘤体积和修订我们以前的肾癌疾病进展模型来获取每个患者的模型的表型-估计的治疗效果。第二步是估计这种表型和传统表型的遗传力，以便了解基因组数据可能解释的患者间变异性的“上限”。第三步是使用OmicKriging方法开发基因组预测工具，第四步(也是最后一步)是在预期的临床试验中验证这些工具。 为了成功地开发基因组预测工具并在 癌症治疗的临床试验，我需要在以下方面的额外培训和经验 基因组学和统计遗传学。我已经找到了两位在这些领域拥有专业知识的导师来监督我的职业发展计划。在他们的指导下，我计划了一套全面的课程和会议，以进行基因组学和统计遗传学的高级培训，以补充我的药物计量学高级学位。有了这个K23奖，我将获得独立开发基因组预测工具的技能，这些工具的表型来自疾病进展模型，并在临床试验中前瞻性地验证这些工具。在未来的工作中，我将演示如何使用这些工具来个性化癌症患者的治疗计划，并最终改善他们的结果。

 DESCRIPTION (provided by applicant): My prior training and long-term career goals make this Translational Scholar Award in Pharmacogenomics and Personalized Medicine (K23) the ideal opportunity to help me become an independent investigator. Within the last three years, I completed joint fellowships in Hematology/Oncology and Clinical Pharmacology and Pharmacogenomics at the University of Chicago. My fellowship research introduced me to the field of pharmacometrics, which is the science of quantifying drug, disease and trial characteristics. Specifically, I became interested i using pharmacometrics to guide the development and optimize the use of cancer therapeutics. I furthered this interest as a Paul Calabresi K12 scholar over the last two years, during which I earned a Master's in Pharmacometrics from one of the only programs in pharmacometrics in this country at the University of Maryland - Baltimore. At the same time, I have gained significant experience writing and conducting clinical trials with major pharmacogenomic components. To advance my career as an independent investigator, I now ask whether pharmacometrics can improve the ability to translate genomic data into clinical decision-making. There have been many pharmacogenomic discoveries describing associations between germline genetic variation and drug toxicity or efficacy. Further discoveries regarding drug efficacy may be enhanced in two ways: first, by identifying better phenotypes of drug efficacy; and second, by utilizing large numbers of common variants in prediction tools rather than relying on a small number of variants. In my fellowship research, we developed a disease progression model of renal cell carcinoma (RCC) that can estimate the treatment effect of drug in the population and in individual patients. This model-estimated treatment effect is an intriguing potential phenotype, as it takes into consideration the full set of longitudinal data regarding tumor size, in contrast to more conventional phenotypes such as objective response (decrease in tumor size by ≥ 30%), progression-free survival and overall survival. Collaborators have shown that semi-automated measurements of tumor volumes, in contrast to the longest diameters on cross-sectional images, might further enhance these phenotypes by providing a more precise assessment of tumor burden. Finally, colleagues of mine at the University of Chicago have developed a method (called OmicKriging) for using all common variants identified during whole genome interrogation (and potentially other -omic data) to make predictions about a phenotype. The COMPARZ trial was the largest ever conducted randomized phase III trial in metastatic RCC, with 1,110 patients randomized to pazopanib versus sunitinib in North America, Europe, and Asia. Approximately two-thirds of these patients provided germline DNA for genome-wide genotyping, which has already been completed. As part of an ongoing collaboration with GlaxoSmithKline Pharmaceuticals (GSK), who sponsored the trial, we have access to clinical data, images from computed tomography scans, and genome-wide genotyping data for these patients. These data offer a unique opportunity to revise our previous disease progression model of RCC using two new therapies and a new phenotype (longitudinal tumor volume), and to explore how common variants can be used to predict both model-estimated treatment effect and conventional phenotypes such as objective response, PFS and OS. The hypothesis is that phenotypes estimated by disease progression models will lead to better genomic prediction tools than conventional phenotypes. These tools could predict which patients are more or less likely to benefit from therapy with tyrosine kinase inhibitors in metastatic RCC. Additionally, these tools could be improved by adding other data elements (such as tumor genotype) and could serve as a blueprint for similar tools using model-based phenotypes in other cancers and other complex diseases. In the research plan, I describe four steps (aims) that logically take us from the raw data to validated genomic prediction tools for both model-based and conventional phenotypes. The first step is to capture the phenotype of model-estimated treatment effect for each patient by measuring tumor volumes and revising our previous disease progression model of RCC. The second step is to estimate the heritability of this phenotype and the conventional ones, in order to understand the "upper limit" of interpatient variability that might be accounted for by genomic data. The third step is to develop the genomic prediction tools using the OmicKriging approach, and the fourth (and final) step is to validate these tools in a prospective clinical trial. In order to be successful in developing genomic prediction tools and prospectively validating them in clinical trials with cancer therapeutics, I need additional training and experience in the areas of genomics and statistical genetics. I have identified two mentors with expertise in these areas to oversee my career development plan. With their guidance, I have planned a comprehensive curriculum of courses and meetings for advanced training in genomics and statistical genetics to supplement my advanced degree in pharmacometrics. With this K23 award, I will acquire the skills to independently develop genomic prediction tools with phenotypes derived from disease progression models and prospectively validate these tools in clinical trials. In future work, I wil demonstrate how these tools can be used to personalize the treatment plan for cancer patients and ultimately improve their outcomes.