New Methods For NMR Metabolic Profiling Normalization And Fusion

High-throughput modern analytical techniques bring a great challenge to subsequent data analysis and interpretation while providing rich information on biological samples.In this paper,metabolomics data analysis is served as the main line.New methods and research ideas were proposed in the following two aspects:normalization method in the data preprocessing stage and the integrated analysis of multivariate statistical analysis modeling phase,which was applied in the study of"Rat model Metabolic Response of Electro-acupuncture Stimulations".Finally,a practical NMR spectral peak assignment software was also designed.The main research results are presented as follows:The first part presented a new method of normalization.The method firstly eliminates variables with large variance based on the relative standard deviations,and then calculates the overall concentration scaling factor by combining the Probabilistic Quotient Normalization method.By using the simulated data and real urine data from"Rat Model Metabolic Response of Electro-acupuncture Stimulations",the new method was compared with the Constant Sum Normalization method and Probabilistic Quotient Normalization method.The results showed that the new method had a more obvious advantage than that of the conventional method under the circumstances of having more different variables,larger differences between groups and greater noise.In the second part,a new method of nuclear magnetic resonance metabolomics data fusion analysis was put forward based on the method of multi-block partial least squares,and the research idea was applied to "Rat Model Metabolic Response of Electro-acupuncture Stimulations".Firstly,multi-block partial least squares integration analysis was used as a multi-block linear regression method.The modeling results were observed by pattern recognition figures and model parameters.The importance of different types of samples was quantitatively weighted by weight parameter.The correlation of the sample variables was analyzed by model's loadings.Secondly,multi-block partial least squares was used as a new integrated filtering method to filter out interference information and then build PLS-DA model,which improved the accuracy of the analysis results.Finally,integrated analysis of multi-block partial least squares was performed based on PLS-DA filtering,which was compared with the non-filtering integration analysis model.The results showed that the integrated analysis model was closer to the conclusion of regression analysis,which showed that consolidating the analysis with the appropriate filtering preprocessing would improve the model's interpretability.The new method of normalization proposed in this paper has good performance and universality,and the method and idea of integration analysis based on multi-block partial least squares algorithm can provide flexible and practical data fusion analysis technology for metabolomics research.Finally,a green,practical,reliable spectral peak assignment software is designed.