Nuclear magnetic resonance combined with genetic algorithm with linear discriminant analysis (GA-LDA) is a suitable model for discriminating urinary metabolomic profiles of individuals with glycemic disorders

Metabolomics is essential in identifying biomarkers involved in the development and progression of diabetes. The study aimed to verify the discrimination among metabolites in the urine of individuals with glycemic alterations using Nuclear Magnetic Resonance (¹H NMR) and multivariate analysis. The preliminary case-control study was performed with three groups: patients with type 2 diabetes (T2D, <i>n</i> = 16), patients with prediabetes (PD, <i>n</i> = 12), and control group individuals (C, <i>n</i> = 11). We obtained the 24-h urine spectra using ¹H NMR from 39 participants. The data were analyzed using Principal Component Analysis (PCA) and supervised analyses. We identified characteristic signals of various metabolites by comparing the chemical shift data with the literature. Our analysis revealed twenty-one distinct metabolic regions, emphasizing citrate, creatinine, glucose, urea, acetate, and glycine. The most pronounced differences were observed in individuals with T2D compared to groups C and PD. We evaluated a range of algorithms to determine the optimal model. The Genetic Algorithm with Linear Discriminant Analysis (GA-LDA) model exhibited remarkable accuracy, sensitivity, and specificity rates of 100% in discriminating between the studied groups. The ¹H NMR and the GA-LDA model are promising methods for discriminating urine metabolomic profiles in case-control studies involving individuals with PD and T2D.

代谢组学（Metabolomics）在鉴定参与糖尿病发生与进展的生物标志物方面具有关键作用。本研究旨在借助核磁共振（Nuclear Magnetic Resonance，¹H NMR）与多变量分析技术，鉴别血糖异常个体尿液中的代谢物差异。本初步病例对照研究共设置三组受试者：2型糖尿病（T2D，n=16）患者、糖尿病前期（PD，n=12）患者，以及健康对照组（C，n=11）个体。研究共纳入39名受试者，对其24小时尿液样本开展¹H NMR光谱检测。采用主成分分析（Principal Component Analysis，PCA）与有监督分析方法对数据进行分析。通过将化学位移数据与已有文献比对，我们鉴定出多种代谢物的特征信号峰。本次分析共识别出21个明确的代谢特征区域，其中柠檬酸、肌酐、葡萄糖、尿素、乙酸盐及甘氨酸最为突出。与健康对照组C及糖尿病前期组PD相比，2型糖尿病组T2D受试者的代谢谱差异最为显著。为筛选最优分类模型，我们对多种算法进行了评估。结合线性判别分析的遗传算法（Genetic Algorithm with Linear Discriminant Analysis，GA-LDA）模型在区分本研究各组受试者时，展现出100%的准确率、灵敏度与特异度。¹H NMR技术与GA-LDA模型可作为区分糖尿病前期及2型糖尿病患者尿液代谢谱的有效方法，在相关病例对照研究中具有良好应用前景。