A New Data Processing Method for Metabonomic and Its Application in a Study of Diabetes
代谢组学数据分析方法及在糖尿病研究中的应用

Multivariate statistical methods are frequently used in nuclear magnetic resonance(NMR)-based metabonomic researches to analyze NMR spectra of biofluids.Based on the fact that the NMR spectrum of a given sample are a sum of the NMR signals from all constituting ingredients,we developed a non-negative matrix factorization(NMF) method,capable of finding parts-based and linear representations of non-negative data,for analyzing the data acquired in NMR-based metabonomic studies.Detail comparisons were made between the NMF method and the commonly use principal component analysis(PCA) method by employing the two methods to discriminate the urine and serum spectra of type-2 diabetic patients from those of the healthy controls.It was shown that,compared to the PCA method,the NMF method is a more effective and accurate method for processing NMR spectra acquired in the metabonomic studies,partially due to its unique features such as the non-negative constraints and part-based representation.The disadvantages of the PCA method were also analyzed and discussed.