核磁共振代谢组学数据的尺度归一化新方法

New Variable Scaling Method for NMR-based Metabolomics Data Analysis

提出了一种新的尺度归一化方法,该方法不强调各变量在尺度上的归一,而是在原始数据的基础上,通过提高稳定性高且在不同类别样本中具有显著差异性的变量的权重,以增强与特征代谢物相关的信息.分别采用模拟数据和真实代谢组学数据对新归一化方法的性能进行评估,并与单位方差法(Unit variance)、变量稳定性(Variable stability)和尺度缩放法(Level scaling)等常用的尺度归一化方法进行了比较.研究结果表明,新归一化方法能够提高多变量统计模型的预测能力,较好地保留了核磁共振谱的分子信息,有助于特征代谢物的识别,并使后续的数据分析结果具有更好的可解释性.

Variable scaling is an important data pre-processing step in NMR metabolomics,especially for biomarkers identification.It aims to make the subsequent multivariate analysis more reliable and easier by highlighting the biomarkers-related variables,and reducing the contamination of the noise and irrelevant variables.A new scaling method is proposed in this paper.The proposed method adjusts the weight of variables by their significance and stabilities in order to enhance the variable probably related to signature metabolites.Both of simulated dataset and real metabolomic dataset are used to estimate the performance of the proposed method.Comparing with Unit variance（UV）,Variable stability（VAST） and Level scaling（LS） methods,the new scaling method would be robust to preserve molecular information of NMR spectra,improving the predictive {ability} of multivariate statistical model and making the results of subsequent analysis more interpretable.Therefore,the method proposed herein is more suitable for biomarker identification.