{"title":"Self-organizing maps for exploration and classification of nuclear magnetic resonance spectra for untargeted metabolomics of breast cancer","authors":"","doi":"10.1016/j.jpba.2024.116377","DOIUrl":null,"url":null,"abstract":"<div><p>Metabolomics has emerged as a powerful tool for identifying biomarkers of disease, and nuclear magnetic resonance (NMR) spectroscopy allows for the simultaneous detection of a wide range of metabolites. However, due to complex interactions within metabolic networks, metabolites often exhibit high correlation and collinearity. To address this challenge, self-organizing maps (SOMs) of Kohonen maps and counter propagation-artificial neural networks (CP-ANN) were employed in this study to model proton nuclear magnetic resonance spectroscopic (<sup>1</sup>HNMR) data from control samples and breast cancer (BC) patients. Blood serum samples from a control group (n=24) and BC patients (n=18) were used to extract metabolites using methanol and chloroform solvents in optimum extraction conditions. The <sup>1</sup>HNMR data was preprocessed by performing phase, baseline, and shift corrections. Subsequently, the preprocessed data was modeled using Kohonen network as an unsupervised technique and CP-ANN as a supervised technique. In this regard, the model built with CP-ANN successfully distinguished between the two classes with an accuracy of 100 % for both group and sensitivity of 96 % and 100 % for control group and BC patients, respectively. Additionally, CP-ANN algorithm demonstrated predictive capabilities by accurately classifying test samples with 90 % sensitivity, 98 % specificity, and 96 % accuracy for control group and 100 % sensitivity, 90 % specificity, and 96 % accuracy for BC patients. Furthermore, analysis of the resulting topological map revealed 14 significant variables (biomarkers) such as sarcosine, lysine, trehalose, tryptophan, and betaine that effectively differentiated between healthy individuals and BC patients.</p></div>","PeriodicalId":16685,"journal":{"name":"Journal of pharmaceutical and biomedical analysis","volume":null,"pages":null},"PeriodicalIF":3.1000,"publicationDate":"2024-07-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of pharmaceutical and biomedical analysis","FirstCategoryId":"3","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0731708524004175","RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"CHEMISTRY, ANALYTICAL","Score":null,"Total":0}
引用次数: 0
Abstract
Metabolomics has emerged as a powerful tool for identifying biomarkers of disease, and nuclear magnetic resonance (NMR) spectroscopy allows for the simultaneous detection of a wide range of metabolites. However, due to complex interactions within metabolic networks, metabolites often exhibit high correlation and collinearity. To address this challenge, self-organizing maps (SOMs) of Kohonen maps and counter propagation-artificial neural networks (CP-ANN) were employed in this study to model proton nuclear magnetic resonance spectroscopic (1HNMR) data from control samples and breast cancer (BC) patients. Blood serum samples from a control group (n=24) and BC patients (n=18) were used to extract metabolites using methanol and chloroform solvents in optimum extraction conditions. The 1HNMR data was preprocessed by performing phase, baseline, and shift corrections. Subsequently, the preprocessed data was modeled using Kohonen network as an unsupervised technique and CP-ANN as a supervised technique. In this regard, the model built with CP-ANN successfully distinguished between the two classes with an accuracy of 100 % for both group and sensitivity of 96 % and 100 % for control group and BC patients, respectively. Additionally, CP-ANN algorithm demonstrated predictive capabilities by accurately classifying test samples with 90 % sensitivity, 98 % specificity, and 96 % accuracy for control group and 100 % sensitivity, 90 % specificity, and 96 % accuracy for BC patients. Furthermore, analysis of the resulting topological map revealed 14 significant variables (biomarkers) such as sarcosine, lysine, trehalose, tryptophan, and betaine that effectively differentiated between healthy individuals and BC patients.
期刊介绍:
This journal is an international medium directed towards the needs of academic, clinical, government and industrial analysis by publishing original research reports and critical reviews on pharmaceutical and biomedical analysis. It covers the interdisciplinary aspects of analysis in the pharmaceutical, biomedical and clinical sciences, including developments in analytical methodology, instrumentation, computation and interpretation. Submissions on novel applications focusing on drug purity and stability studies, pharmacokinetics, therapeutic monitoring, metabolic profiling; drug-related aspects of analytical biochemistry and forensic toxicology; quality assurance in the pharmaceutical industry are also welcome.
Studies from areas of well established and poorly selective methods, such as UV-VIS spectrophotometry (including derivative and multi-wavelength measurements), basic electroanalytical (potentiometric, polarographic and voltammetric) methods, fluorimetry, flow-injection analysis, etc. are accepted for publication in exceptional cases only, if a unique and substantial advantage over presently known systems is demonstrated. The same applies to the assay of simple drug formulations by any kind of methods and the determination of drugs in biological samples based merely on spiked samples. Drug purity/stability studies should contain information on the structure elucidation of the impurities/degradants.