Metabolic Patterns in 1H NMR Spectra of Biofluids (RMI)

描述（由申请人提供）：1H核磁共振（NMR）广泛用于研究生物样品的代谢状态。其非侵入性的性质和检测多种化合物的能力使其能够随着时间的推移以高水平的细节跟踪复杂的生化过程，从而导致其在“代谢组学”或表征生命系统对病理生理刺激或遗传修饰的代谢物反应中的用途。它从数百个相关样品中产生具有大量共振的光谱。模式识别技术的使用受到两个因素的限制：1）由温度，pH值等的微小差异引起的频率的微小系统变化，以及2）使用主成分分析（PCA）来识别数据集中变化的数学成分，而不是动态变化的真正物理来源。频率的变化导致将光谱“分箱”到只有几百个点，而使用PCA来识别数据集中的潜在模式使得很难找到物理上有意义的代谢模式。在这个R33中，我们通过对高分辨率NMR数据进行稳健的预处理，然后应用贝叶斯谱分解（BSD）和约束非负矩阵分解（cNMF）来揭示描述系统变化的潜在代谢模式，从而解决了这些问题。有五个主要的具体目标 具体目标一：为尿液的一系列高分辨率1H NMR光谱开发半自动化预处理程序，以提高光谱质量，使PR程序能够识别任何潜在的生化相关光谱模式。 具体目标二：实现贝叶斯谱分解（BSD）作为一个实用的易于使用的频谱分析程序。 具体目标三。将约束非负矩阵分解（cNMF）作为一种实用易用的频谱分析程序 具体目标四。在Linux集群上以并行代码的形式实现BSD。 具体目标五：应用BSD和cNMF分析从大鼠和小鼠毒理学研究中获得的一系列经具体目标一技术预处理的尿液核磁共振谱。

DESCRIPTION (provided by applicant): 1H nuclear magnetic resonance (NMR) is widely used to investigate the metabolic state of biological samples. Its non-invasive nature and ability to detect multiple compounds allow it to follow complex biochemical processes over time at a high level of detail leading to its use in "metabolomics" or characterization of the metabolite response of living systems to pathophysiological stimuli or genetic modification. It generates spectra with a large number of resonances from hundreds of related samples. The use of pattern recognition techniques here have been restricted by two factors: 1) small systematic variations in frequency caused by small differences in temperature, pH, etc. and 2) use of principal component analysis (PCA) for the identification of the mathematical components of the variation in the dataset, rather then true physical sources for dynamic changes. The variations in frequency have led to "binning" the spectra to only a few hundred points while the use of PCA to identify underlying patterns in the datasets makes finding physically meaningful metabolic patterns hard. In this R33 we address these problems by robust pretreatment of high resolution NMR data and then applying Bayesian Spectral Decomposition (BSD) and constrained Non-negative Matrix Factorization (cNMF) to uncover the underlying metabolic patterns that describe the change in the system. There are five main Specific Aims Specific Aim I: Develop a semi-automated preprocessing procedure for series of high resolution 1H NMR spectra of urine to improve spectral quality to enable PR procedures to identify any underlying biochemically relevant spectral patterns. Specific Aim II: Implement Bayesian Spectral Decomposition (BSD) as a practical easy-to-use spectral analysis procedure. Specific Aim III. Implement constrained Non-negative Matrix Factorization (cNMF) as a practical easy-to-use spectral analysis procedure Specific Aim IV. Implement BSD as parallel code on a Linux cluster. Specific Aim V. Apply BSD and cNMF to analyze a series of NMR spectra of urine acquired from toxicology studies of rats and mice that have been preprocessed by the techniques of Specific Aim I.