Journal Article Epublish
PubMed: 40379752 DOI: 10.1038/s41598-025-01895-2
Metabolomics and machine learning identify urine metabolic characteristics and potential biomarkers for severe Mycoplasma pneumoniae pneumonia.
Sci Rep 15(1):17090
Authors: Lin Li, Wang Haijun, Chen Tianyu, Wang Wenjing, Wang Zi, Cheng Yibing
Affiliations
- Department of Emergency, Henan Province Engineering Research Center of Diagnosis and Treatment of Pediatric Infection and Critical Care, Children's Hospital Affiliated to Zhengzhou University, Zhengzhou, 450018, China.
- JIUHE Diagnostics Co. Ltd, Zhengzhou, 450016, China.
- Department of Emergency, Henan Province Engineering Research Center of Diagnosis and Treatment of Pediatric Infection and Critical Care, Children's Hospital Affiliated to Zhengzhou University, Zhengzhou, 450018, China.
Abstract
To study the differences in the urine metabolome between pediatric patients with severe Mycoplasma pneumoniae pneumonia (SMPP) and those with general Mycoplasma pneumoniae pneumonia (GMPP) via non-targeted metabolomics method, and potential biomarkers were explored through machine learning (ML) algorithms. The urine metabonomics data of 48 children with SMPP and 85 children with GMPP were collected via high performance liquid chromatography‒mass spectrometry (HPLC-MS/MS). The differential metabolites between the two groups were obtained via principal component analysis (PCA) and partial least squares-discriminant analysis (PLS-DA), and the significant metabolic pathways were screened via enrichment analysis. Potential biomarkers were identified using the random forest algorithm, and their relationships with clinical indicators were subsequently analyzed. A total of 136 significantly differential metabolites were identified in the urine samples from SMPP and GMPP. Of these, 68 metabolites were upregulated, and 68 were downregulated, predominantly belonging to the amino acid group. A total of 6 differential metabolic pathways were enriched, including Galactose metabolism, Pantothenate and CoA biosynthesis, Cysteine and methionine metabolism, Biotin metabolism, Glycine, serine and threonine metabolism, Arginine biosynthesis. Three significant potential biomarkers were identified through machine learning: 3-Hydroxyanthranilic acid (3-HAA), L-Kynurenine, and 16(R)-HETE. The area under the receiver operating characteristic curve (AUC) for this three-metabolite panel was 0.9142. There are great differences in the urine metabolome between SMPP and GMPP children, with multiple metabolic pathways being abnormally expressed. Three metabolites have been identified as potential biomarkers for the early detection of SMPP.
MeSH
Algorithms, Biomarkers, Machine Learning, Pneumonia, Metabolomics, Learning, Mycoplasma, Mycoplasma Pneumoniae, Child