Email:
support-global@metwarebio.com
Aug 01, 2023
We are very pleased to share the chronic kidney disease study published in Chinese Chemical Letters Journal. The authors of 'Metabolome profiling by widely-targeted metabolomics and biomarker panel selection using machine-learning for patients in different stages of chronic kidney disease (doi.org/10.1016/j.cclet.2024.109627)' utilized our proprietary the TM widely-targeted metabolomics detection and reported 1431 metabolites. The purpose of this research is to develop a new method for diagnosing chronic kidney disease (CKD) by identifying specific metabolites in blood plasma. Current methods for diagnosing CKD are not very accurate, but this new method, which combines targeted and untargeted metabolomics technologies, has the potential to be more accurate. The researchers will use machine learning to identify the most important metabolites for diagnosing different stages of CKD. This new method could be especially helpful for people in Asia, where CKD is a major public health problem.
The study included 83 participants, divided into healthy controls and CKD patients at different stages. The researchers developed a new method (TM widely-targeted metabolomics) to identify metabolites in blood plasma samples. They used a combination of high-resolution and triple quadrupole mass spectrometry to analyze the samples. A total of 1431 compounds were identified, with high confidence levels for most of them. The results showed good reproducibility and reliability. The researchers then removed exogenous compounds from the data set for further analysis.
Through cluster analysis, researchers divided 539 endogenous metabolites into six groups based on their trends during progression from healthy controls to the fifth stage of chronic kidney disease (CKD). Some metabolites exhibited uni-directional changes, increasing or decreasing with the progression of renal impairment, while others showed different patterns of change. Among these endogenous metabolites, 74.03% showed significant differences among the four groups, indicating widespread metabolic changes accompanying CKD progression. Levels of 99 metabolites were upregulated, 63 metabolites were downregulated, and the remaining 377 metabolites exhibited non-linear changes. The most affected metabolites included amino acids, organic acids, glycerophospholipids, nucleotides, and their metabolites.
Further analysis focused on the top 30 significantly altered metabolites, categorized by type. Results revealed significant changes in amino acids and their metabolites, nucleotides and their metabolites, as well as organic acids and their derivatives during CKD progression. Correlation analysis identified 5-aminoimidazole ribonucleotide, aspartyl glutamine, and pyrroloquinoline quinone alcohol as having the highest total correlation coefficients, indicating close relationships among these metabolites. Through KEGG pathway analysis, four significantly altered pathways with FDR < 0.05 related to amino acid metabolism were discovered. These findings highlight specific categories of metabolites undergoing significant changes in CKD and underscore alterations in metabolic pathways associated with the disease.
The study employed machine learning techniques to identify the optimal combination of metabolites for predicting different stages of CKD. Utilizing a random forest model for feature selection, 7 most important metabolites were selected from an initial pool of 539 endogenous metabolites. Various machine learning models were then fitted to these metabolites, and optimal parameters were determined through grid search and cross-validation. Based on high accuracy and f1 score, the random forest model was ultimately chosen. This optimal model effectively predicted CKD stages in both the testing and training datasets.
The study validated the performance of the selected biomarker combination by comparing it with three other metabolite combinations: top 10 upregulated metabolites, top 10 downregulated metabolites, and top 10 metabolites with nonlinear changes. Each metabolite combination was fitted using a random forest model, and receiver operating characteristic (ROC) curves were plotted in the testing dataset. The area under the ROC curve (AUC) was calculated for each metabolite combination, with the 7-metabolite model exhibiting higher performance compared to the other three combinations. UMAP analysis confirmed the clustering ability of each metabolite combination. Both supervised and unsupervised machine learning techniques demonstrated the superior performance of the 7-metabolite combination. The results also highlighted the level changes of these 7 metabolites across different CKD stages, emphasizing their potential importance and their association with metabolic pathways and cellular processes related to renal dysfunction in CKD. Overall, these findings underscore the robustness and accuracy of the 7-metabolite combination in predicting different CKD stages, with 5-aminoimidazole-4-carboxamide ribonucleotide emerging as the most important feature among these 7 metabolites.
In summary, the study presents an efficient and sensitive TM widely-targeted metabolomics approach capable of comprehensively, sensitively, and accurately detecting and quantifying metabolites, thus providing deeper insights into disease-related metabolic changes. Utilizing this technique, they identified 1431 compounds from plasma samples and, focusing on 539 endogenous metabolites, employed machine learning algorithms to identify 7 metabolites capable of accurately distinguishing between different stages of chronic kidney disease (CKD) patients. The discovery of these 7 metabolites as potential diagnostic biomarkers and therapeutic targets holds promise for improving the management of CKD.
Discover comprehensive metabolomics, lipidomics, and proteomics solutions at Metware Lab in Boston. Enjoy a 10% discount on our targeted metabolomics services from April 15th to May 31st, 2024. Contact us today to accelerate your research!
