Efficient Discovery of Medical Associations

描述（由申请人提供）：发现影响人类健康的关键因素在生物医学研究中非常重要，可以为公众提供有价值的健康指导。随着获取和存储医疗数据能力的增强，对促进此类发现的计算工具的需求越来越大，而且还在不断增加。本项目旨在建立一个有效的风险因素/关联发现系统，从医疗数据集中提取重要的、有效的、非冗余的和以前未知的属性关联。具体目标是：具体目标1：创建一个有效的关联发现算法，集成基于知识和客观关联挖掘技术的优势。在初步研究中，我们设计了一种新的基于知识的关联分析算法，该算法可以检测和丢弃大多数无效或已知的关联。为了进一步提高分析质量和剔除冗余关联，我们建议充分研究基于知识的算法，并将其与客观冗余减少技术MAFIA相结合。这种新的关联发现算法将只呈现非平凡的、有效的、非冗余的和以前未知的关联，这些关联可以导致令人兴奋的发现。具体目标2：研究一种新的基于语义网络的知识模型。在关联分析中系统地应用用户和医学领域知识是最近才开始的。Chen等人人工构建了由衷心青少年健康调查的38个属性组成的语义网络，成功识别出青少年健康与发展的8个关联。为了提高语义网络构建的效率，本项目将开发一个自动语义网络构建组件，直接从NIH开发的大型医学知识库“统一医学语言系统”中提取知识。为了验证和评估该模型的可扩展性，将自动构建一个基于大型真实医疗数据集的大型语义网络并进行全面测试。具体目标3：严格的评估

DESCRIPTION (provided by applicant): Discovery of critical factors affecting human health is very important in biomedical research, and can provide valuable health guidance to the general public. As the capability to capture and store medical data expands, the need for computational tools that facilitate such discoveries is high and increasing. This project aims to build an efficint risk factor/association discovery system to extract significant, valid, non-redundant, and previously unknown associations of attributes from medical datasets. The specific aims are: Specific Aim 1: Creating an efficient association discovery algorithm integrating the strengths of both knowledge-based and objective association mining techniques. In the preliminary study we designed a novel knowledge-based association analysis algorithm, which can detect and discard the majority of invalid or already known associations. To further improve the analysis quality and weed out redundant associations, we propose to fully investigate our knowledge-based algorithm and integrate it with MAFIA, an objective redundancy-reducing technique. This new association discovery algorithm will present only non-trivial, valid, non-redundant, and previously unknown associations that can lead to exciting discoveries. Specific Aim 2: Investigating a novel semantic network-based knowledge model. Systematic application of user and medical domain knowledge in association analysis started only recently. Chen et al. manually built a semantic network consisting of 38 attributes of the Heartfelt adolescent health survey and successfully identified 8 associations on adolescent health and development. To improve the efficiency of semantic network construction, the proposed project will develop an automatic semantic network building component to extract knowledge directly from the Unified Medical Language System, a large medical knowledge base developed by NIH. To validate and evaluate the scalability of this model, a large semantic network based on large real-world medical datasets will be automatically built and fully tested. Specific Aim 3: Rigorous evaluation will be conducted with a thorough analysis of large real- world datasets, and discovered associations will be validated with biostatistic models and experienced medical researchers. Since one advantage of our association discovery technique is the capability of finding "hidden" associations that a user has never suspected, new health-associated factors may be discovered, which will undoubtedly advance the public health research. National Institute of General Medical Sciences (NIGMS) commits to investing in discovery by using a variety of vehicles to support basic research. This project aims to efficiently discover critical factors associated with human health, which directly contribute to enhance the basic biomedical research and support the creation of research resources including software and hardware tools (Goal 1 and 2 in the NIGMS Strategic Plan). Our highly experienced and interdisciplinary research team consists of 1 Computer Scientist, 1 computer technician/programmer, 1 Bio-statistician, and 2 medical technicians, and naturally this project will advance multidisciplinary and interdisciplinary inquiry (Goal 2). The University of Houston-Downtown is both a Hispanic Serving Institute (HSI) and Minority Institute (MI) designated by the U.S. Department of Education, and this project will directly involve underrepresented minority students into biomedical research. More students will benefit from class projects and new courseware spawned from this project (Goal 3 in the NIGMS Strategic Plan).