Transfer Rule Learning with Functional Mapping for Integrative Modeling of Panomics Data

 描述(申请人提供)：旨在早期发现和更好地管理癌症等疾病的科学研究中的分子图谱数据，积累的速度远远超出了我们有效提取对精确医学实践有价值的知识的能力。一个主要的挑战是，这些数据通常是使用多种高通量技术产生的，从而产生相同或相关分类任务的泛组学数据，如基因表达和DNA甲基化。该项目将开发迫切需要的计算方法和工具，用于对泛组学数据进行综合建模，以改进相关分子图谱研究中的疾病状态分类。该项目将扩展以前开发的转移规则学习(TRL)方法，通过自动从一个数据集中学习分类规则，转移知识并在从相关数据集中学习规则时使用，来处理生物标志物剖析研究中的稀疏数据。该项目将开发、应用和评估一种新的知识转移方法，该方法涉及使用本体论或分类层次结构以及分类规则学习。具体地说，这个项目将测试这样一个假设，即通过本体结构使用功能映射(TRL-FM)传递分类规则的学习，以提供特定于领域的相关性，比传统方法改进泛组学数据的综合建模，以产生更好的预测性能，并为疾病状态分类识别更稳健的生物标志物面板。TRL-FM原型将应用于来自两个不同领域的现有去识别泛组学数据，用于(1)癌症和(2)使用微生物组图谱的全球人群中的寄生虫感染的分类。TRL-FM模型将利用一组来自一组高风险CT筛查患者的现有的去识别的泛组学数据和相关的结节大小信息，对精确的肺癌分类和稳健的生物标记物发现进行验证，并与最先进的分类器进行全面比较。该项目可以帮助创建更强大的筛查工具，对肺癌进行精确分类，肺癌是美国主要的癌症死亡原因。该项目还将影响全球健康，有可能 利用从粪便微生物组图谱中获得的数据，帮助改进由蠕虫引起的感染的筛查和管理，蠕虫是影响全球10多亿人的最常见寄生虫。该项目将产生计算工具，当从泛组学数据建立预测模型时，可以有效地整合来自多个来源的知识。预测模型是高度可解释的，捕获了数据中子总体下的模式，作为分类规则，增加了关于区别性生物标志物稳健性的信息。该项目将创建工具，通过纳入从核糖体RNA测序的细菌物种分析的专门知识，使快速增长的人类微生物群研究界受益。TRL-FM工具将使微生物组数据的综合建模更有效率，从而能够快速洞察危害或支持人类健康的细菌菌株和物种。

 DESCRIPTION (provided by applicant): Molecular profiling data from scientific studies aiming for early detection and better management of diseases such as cancer, has accumulated at rates far beyond our abilities to efficiently extract knowledge of value to the practice of precisin medicine. A major challenge is that these data are often generated using multiple high-throughput technologies giving rise to panomics data such as gene expression and DNA methylation for the same or related classification task. This project will develop critically neede computational methods and tools for the integrative modeling of panomics data to improve disease state classification from related molecular profiling studies. This project will extend Transfer Rule Learning (TRL) methods that were previously developed to deal with sparse data from biomarker profiling studies, by automatically learning classification rules from one dataset, transferring that knowledge and using it when learning rules from a related dataset. This project will develop, apply and evaluate a novel method for knowledge transfer that involves the use of ontological or taxonomic hierarchies along with classification rule learning. Specifically, this project will test the hypothesis that transfer learning of classification rules using functional mapping (TRL-FM) via ontological structure to provide domain-specific relatedness improves integrative modeling of panomics data over conventional methods to yield better predictive performance and identify more robust biomarker panels for disease state classification. The TRL-FM prototype will be applied to existing de-identified panomics data from two diverse domains for the classification of (1) cancer, and (2) parasitic infections in global populations using microbiome profiling. The TRL-FM models will be validated for precise lung cancer classification and robust biomarker discovery, using an existing set of de-identified panomics data and related nodule size information from a cohort of high-risk CT-screened patients, and comprehensively compared to state-of-the-art classifiers. This project can help create more robust screening tools for the precise classification of lung cancer, the leading cause of death from cancer in the United States. This project will also impact global health with the potential to help improve screening and management of infections caused by helminths, the most common parasites affecting more than a billion people worldwide, using data obtained from fecal microbiome profiling. This project will result in computational tools that can efficiently integrat knowledge from multiple sources when building predictive models from panomics data. The predictive models are highly interpretable, capturing patterns that underlie subpopulations in the data, as classification rules with augmented information about the robustness of discriminative biomarkers. This project will create tools to benefit the rapidly growing human microbiome research community, by incorporating knowledge specific to the analyses of bacterial species sequenced from ribosomal RNA. The TRL-FM tools will make integrative modeling of microbiome data more efficient thereby enabling rapid insights into bacterial strains and species that harm or support human health.