Plasma markers for gastric cancer diagnosis: a metabolomics- and machine learning-based exploratory study

  Objective  To investigate the significance of plasma metabolites for gastric cancer diagnosis based on metabolomics and machine learning algorithms.

  Methods  Plasma samples were collected from 20 gastric cancer patients and 20 gender- and age-matched healthy volunteers (controls). After extracted with methanol, the metabolites in the plasma samples were analyzed with chromatography-mass spectrometry and annotated with mzCloud, mzVault and Masslist databases. The differential metabolites between the cases and controls were screened with the value of variable importance of projection (VIP: > 1, P < 0.05, log₂|fold change| > 1); the identified differential metabolites were subjected to the Kyoto Encyclopedia of Genes and Genomes (KEGG) database enrichment using hypergeometric tests to detect differential metabolic pathways between the two groups. Specific metabolite-based diagnostic model and receiver operating characteristic (ROC) curve were established using Boruta algorithm to identify significant differential metabolites for gastric cancer diagnosis. Relative contents of the differential metabolites were also calculated.

  Results  A total of 230 differential metabolites were screened out from the plasma extracts of the two groups and 5 differential metabolic pathways were identified, including phenylalanine metabolism, biosynthesis of phenylalanine/tyrosine/tryptophan, arginine biosynthesis, histidine metabolism, and pantothenic acid/coenzyme A biosynthesis. The identified 9 significant differential metabolites for gastric cancer diagnosis were L-aspartic acid, 2-phenylethylamine, ornithine, hippuric acid, citrulline, pantetheine, 1-methylhistamine, L-tyrosine, and L-histidine and the relative levels of all the 9 differential metabolites were significantly lower in the plasma from the cases than those from the controls (P < 0.05 for all).

  Conclusion  The enriched five metabolic differential pathways may be involved in the development of gastric cancer, and the nine differential metabolites can be used as metabolic biomarkers for gastric cancer diagnosis. The combination of metabolomics and machine learning algorithms could help identify markers for gastric cancer diagnosis.