Machine Learning Tools for Discovery and Analysis of Active Metabolic Pathways

 描述(由申请人提供)：该项目旨在为来自不同平台的代谢组学数据开发新的统计机器学习方法，包括用于测量代谢flUX和核磁共振平台的目标和无偏/全球质谱学(MS)、标记MS实验。无偏的MS和核磁共振fi研究导致识别出大量未命名的光谱，这些光谱不能与已知的代谢物直接匹配，因此经常在下游分析中被丢弃。fi的第一个目标是开发一种新的核惩罚回归方法，用于分析无偏规划研究的数据。它提供了一个系统的框架，用于从 未命名的光谱通过一个内核突出了样品之间的相似和不同之处，并反过来增强了来自已命名代谢物的信号。这导致了与感兴趣的表型相关的命名代谢物的识别fi阳离子能力的提高，以及预测准确性的提高。还建议对这个基于内核的框架进行扩展，以允许系统地集成来自不同PROfi研究的代谢组学数据，例如有针对性和无偏见的MS PROfi技术。第二个目标为核惩罚回归提供了一个正式的推理框架，从而补充了fi第一个目标的发现阶段。第三个目标是代谢途径浓缩分析，它既测试给定途径中稳态代谢物活性的协调变化，也测试代谢反应机制中的异常。该项目的第四个目标是提供一个统一的框架，用于基于网络的静态(基于质谱学)和动态(基于代谢fiUX)代谢组学测量的综合分析，从而提供代谢组和fluxome的综合视图。最后，利用R语言、Cytosscape平台和Galaxy Workflow系统的功能，在易于使用的开源软件中实现所提出的方法，从而为代谢组学领域的进一步发展提供一个可扩展的平台。拟议的软件工具还将提供数据存储库和协调中心(DRCC)数据集的插件，NIH共同基金代谢组学计划支持的所有区域代谢学中心都将存储经过管理的数据。

 DESCRIPTION (provided by applicant): This project aims to develop new statistical machine learning methods for metabolomics data from diverse platforms, including targeted and unbiased/global mass spectrometry (MS), labeled MS experiments for measuring metabolic flux and Nuclear Magnetic Resonance (NMR) platforms. Unbiased MS and NMR profiling studies result in identifying a large number of unnamed spectra, which cannot be directly matched to known metabolites and are hence often discarded in downstream analyses. The first aim develops a novel kernel penalized regression method for analysis of data from unbiased profiling studies. It provides a systematic framework for extracting the relevant information from unnamed spectra through a kernel that highlights the similarities and differences between samples, and in turn boosts the signal from named metabolites. This results in improved power in identification of named metabolites associated with the phenotype of interest, as well as improved prediction accuracy. An extension of this kernel-based framework is also proposed to allow for systematic integration of metabolomics data from diverse profiling studies, e.g. targeted and unbiased MS profiling technologies. The second aim pro- vides a formal inference framework for kernel penalized regression and thus complements the discovery phase of the first aim. The third aim focuses on metabolic pathway enrichment analysis that tests both orchestrated changes in activities of steady state metabolites in a given pathway, as well as aberrations in the mechanisms of metabolic reactions. The fourth aim of the project provides a unified framework for network-based integrative analysis of static (based on mass spectrometry) and dynamic (based on metabolic flux) metabolomics measurements, thus providing an integrated view of the metabolome and the fluxome. Finally, the last aim implements the pro- posed methods in easy-to-use open-source software leveraging the R language, the capabilities of the Cytoscape platform and the Galaxy workflow system, thus providing an expandable platform for further developments in the area of metabolomics. The proposed software tool will also provide a plug-in to the Data Repository and Coordination Center (DRCC) data sets, where all regional metabolomics centers supported by the NIH Common Funds Metabolomics Program deposit curated data.