INFORMATICS METHODS AND MODELS FOR COMPUTATIONAL PATHOLOGY

摘要 申请人是贝斯以色列女执事医疗中心和哈佛医学院的助理教授。 申请人的研究计划侧重于开发翻译生物信息学方法， 异构生物医学数据类型（基因组、转录组、显微镜、临床）， 为患者开发改进的诊断和治疗方法。申请人最近完成的项目 包括：开发用于构建预后模型的计算病理学（C-Path）平台 从显微图像数据（科学转化医学，2011年）;和发展的意义 用于鉴定稳健的预后基因表达标记形式的预后标记分析方法 临床注释的基因组数据（PLoS Computational Biology，2013）。 该K22奖项将为申请人提供必要的支持，以实现以下目标：（1） 成为设计和使用本体建模计算病理学数据并支持 计算病理学知识库;（2）成为综合预测模型的专家， 计算病理学数据、基因表达数据和临床结果数据;以及（3）开发和 管理翻译生物信息学的独立研究生涯。为了实现这些目标，申请人 已经组建了一个由导师和合作者组成的团队，他们在这些翻译领域都有专业知识。 生物医学信息学该团队包括：艾萨克·科汉博士，他是儿科教授，指导 儿童医院信息学计划，并领导一个NLM支持的国家生物医学中心 计算机;哈佛医学院教授罗恩基基尼斯博士和罗伯特格林 布里格姆妇女医院放射科生物医学信息学杰出主任 约翰·夸肯布什博士，他是生物统计学和计算生物学教授， 丹娜-法伯癌症研究所的癌症生物学教授。 在K22奖励计划期间，申请人及其研究团队将开发信息学模型和方法 计算病理学数据。在目标1中，他们将开发一个计算病理学本体，以支持 计算病理学知识库。知识库将填充微观表型 数据，基因表达数据和临床结果数据，来自2,500多名癌症患者， 癌症基因组图谱项目的一部分。在具体目标2中，他们将开发和应用 机器学习中的方法，以确定基因表达和微观表型之间的关联。 这些信息将纳入C-Path知识库。在目标3中，他们将使用C路径 建立综合预测模型的知识库， 定量基因表达数据来预测患者的存活率。综合预后模型产生于 乳腺癌、脑癌、肾癌和肺癌将导致对这些恶性肿瘤更有效的诊断。 该应用的中心假设是形态学和分子数据固有地 互补的，最具生物学信息和临床有用的预测模型将纳入 这两种异构数据类型的信息。这项研究将成为R 01的基础。 应用程序，以进一步开发和验证本项目中开发的信息学方法和模型。

Abstract The applicant is an Assistant Professor at Beth Israel Deaconess Medical Center and Harvard Medical School. The applicant's research program focuses on developing methods in translational bioinformatics to integrate heterogeneous biomedical data types (genomic, transcriptomic, microscopic, clinical) enabling the development of improved diagnostics and therapeutics for patients. Recent projects completed by the applicant include: the development of the Computational Pathology (C-Path) platform for building prognostic models from microscopic image data (Science Translational Medicine, 2011); and the development of the Significance Analysis of Prognostic Signatures method for identifying robust prognostic gene expression signatures form clinically annotated genomic data (PLoS Computational Biology, 2013). This K22 award will provide the applicant with the support necessary to accomplish the following goals: (1) to become an expert at designing and using ontologies to model Computational Pathology data and to support a Computational Pathology Knowledgebase; (2) to become an expert at integrative predictive modeling of Computational Pathology data, gene expression data, and clinical outcomes data; and (3) to develop and manage an independent research career in translational bioinformatics. To achieve these goals, the applicant has assembled a team of mentors and collaborators with expertise in each of these areas of translational biomedical informatics. The team includes: Dr Isaac Kohane, who is a Professor of Pediatrics, directs the Children's Hospital Informatics Program and leads an NLM-supported national center for biomedical computing; Dr Ron Kikinis who is a Professor at Harvard Medical School, and the Robert Greenes Distinguished Director of Biomedical Informatics in the Department of Radiology at Brigham and Women's Hospital; and Dr John Quackenbush, who is a Professor of Biostatistics and Computational Biology and Professor of Cancer Biology at the Dana-Farber Cancer Institute. During the K22 award program, the applicant and his study team will develop informatics models and methods for Computational Pathology data. In Aim 1, they will develop a Computational Pathology Ontology to support a Computational Pathology Knowledgebase. The knowledgebase will be populated with microscopic phenotype data, gene expression data, and clinical outcomes data from over 2,500 cancer patients that underwent molecular profiling as part of The Cancer Genome Atlas project. In Specific Aim 2, they will develop and apply methods in machine learning to identify associations between gene expression and microscopic phenotypes. This information will be incorporated into the C-Path Knowledgebase. In Aim 3, they will use the C-Path Knowledgebase to build integrative prognostic models that jointly model quantitative morphological data and quantitative gene expression data to predict patient survival. The integrative prognostic models generated in breast, brain, kidney, and lung cancer will lead to more effective diagnostics for these malignancies. The central hypothesis for this application is that morphological and molecular data are inherently complementary, and the most biologically informative and clinically useful predictive models will incorporate information from both of these heterogeneous data types. This research will form the basis for an R01 application to further develop and validate the informatics methods and models developed in this project.