Metabolomics: Plasma Panel Detects Early Gastric Cancer
Gastric cancer is usually found late, and the blood tests meant to catch it early keep failing: CEA, CA19-9 and CA72-4 each miss roughly half of confirmed cases. A plasma metabolomics study published in Nature Communications in May 2026 asked whether small-molecule measurement could close that gap, using 1,706 participants from four Chinese centers. Three analytical stages ran in sequence, from untargeted discovery through relative quantitation to absolute quantitation, ending in a 12-metabolite signature the authors named PMB-P12.
Key Takeaways
A 12-metabolite plasma panel derived by metabolomics separated gastric cancer from non-cancer at an AUC of 0.951 in an independent 309-person cohort.
The panel held up in stage IA disease at 92.2 percent sensitivity, the window where endoscopic resection is still curative.
Conventional markers trailed in the same samples: CEA 0.649 AUC, CA19-9 0.541, CA72-4 0.542.
Absolute concentrations, not relative peak areas, let a metabolite panel move between labs without rebuilding its threshold.
Three-Stage Metabolomics Narrows 3,327 Signals to 12
Discovery profiling of 469 subjects on a TripleTOF 6600 and a QTRAP 6500 returned 3,327 metabolic signals. Differential testing across ten pairwise comparisons, then confirmation against authentic standards, cut that to 84 metabolites at MSI Level 1 (the reporting tiers formalized by Sumner and colleagues, 2007). A second cohort of 928 subjects was measured by absolute-quantitative MRM, and a selection routine the authors call BIO-FIRE pulled out the final twelve. The panel is chemically mixed: niacinamide, taurine, sphingosine, hypoxanthine and 12-HETE sit beside monoethylhexyl phthalic acid, a phthalate exposure metabolite rather than anything tumor-intrinsic.
Key Findings
Validation held across centers: random forest scoring gave an AUC of 0.951 (95% CI 0.928 to 0.974), with 0.904 sensitivity and 0.902 specificity in 309 external subjects.
Early-stage sensitivity did not collapse: stage IA cancers were detected at 0.922 sensitivity, tumors under 4 cm at 0.929.
Marker-negative patients were still caught, at 0.948 sensitivity in the subgroup CEA, CA19-9 and CA72-4 had all scored normal.
Benign disease sat in the comparator arm, which makes this a harder test than cancer versus healthy.
Figure 1. Study design and analytical workflow. Panel a shows 1,706 individuals enrolled at four centers into discovery (n=469), modeling (n=928) and validation (n=309) cohorts spanning healthy controls, benign gastric disease, early and advanced gastric cancer. Panel b traces untargeted and relative-quantitative targeted profiling from 3,327 signals down to 84 MSI Level 1 metabolites. Panel c shows absolute-quantitative measurement plus BIO-FIRE feature selection yielding the 12-metabolite panel. Panel d covers model training, internal testing, external validation and four clinical scenarios. Adapted from Bai et al. (2026), Nature Communications.
Reading the Model Numbers Carefully
Eight algorithms were compared, and random forest won. Its training AUC of 1.000 is best read as an ensemble memorizing 696 samples, which is why the 0.921 internal test result and the 0.951 external figure carry the weight. External performance exceeding internal performance is unusual, and probably reflects cohort composition rather than real generalization gain.
Where the Panel Would Actually Be Used
Each scenario tested maps onto a real gap in gastric cancer triage: negative conventional markers, tumors under 4 cm, esophagogastric junction cancers, and stage IA lesions eligible for endoscopic submucosal dissection. This is triage, not population screening, and that distinction shapes how a follow-up trial should be powered. The same logic runs through plasma panels built for multi-cancer detection, where early-stage sensitivity decides whether a test changes management.
From Discovery Panel to Transferable Assay
Moving from relative peak areas to absolute concentrations is the part of this workflow most worth copying, since a panel reported in real units can be re-run on another instrument in another country without re-deriving its cutoff. We see the same pattern at Dalton, where targeted metabolomics assays that skip absolute quantitation stall the moment a project crosses from discovery into validation. For teams scoping multi-omics biomarker discovery, lock chemistry, calibration curves and batch-correction strategy before the cohort grows.
Frequently Asked Questions
What is metabolomics used for in cancer detection?
Metabolomics measures the small molecules made by cells and tissues, which shift faster than genes or proteins once disease alters cellular activity. In cancer detection it finds blood-measurable patterns that separate patients from controls. Sitting at the end of biological cascades, metabolites report a tumor's active state rather than its potential.
How accurate is the metabolomics panel for early gastric cancer?
In a validation cohort of 309 people the 12-metabolite panel reached an AUC of 0.951 overall and 0.922 sensitivity in stage IA disease, well above CEA, CA19-9 and CA72-4 in the same samples. The work is retrospective and drawn from Chinese centers only, so prospective multi-ethnic testing is still needed.
How large does a plasma metabolomics biomarker study need to be?
This one used 1,706 participants split into discovery, modeling and validation groups, which is large for a targeted panel study. Sets of 100 to 200 samples per group are more typical at discovery, but an independent cohort run as a separate batch is what makes a result credible. Cost tracks sample count and LC-MS method depth.
Conclusion
A small, absolutely quantified metabolite panel beating legacy serum markers in a Chinese gastric cancer population is believable. Whether those twelve analytes behave the same way under different diets, genetic backgrounds and a prospective design isn't proven, and the plasticizer metabolite among them sharpens that question. Teams planning similar studies should budget for absolute quantitation early, because that step turns a signature into an assay another lab can run.
Related Reading
See how we run these analyses in one lab: Dalton's multi-omics CRO services.
Citation
Bai, L., Hu, F., Zhang, W., Peng, H., Peng, H., Li, H., Zhu, X., Xie, Y., Zhang, S., & Min, L. (2026). Multi-phase hybrid metabolomics framework identifies clinically applicable plasma signatures for early detection of gastric cancer. Nature Communications, 17, 6372. https://doi.org/10.1038/s41467-026-72983-8
Note
This blog post summarizes findings from the above-cited research. Figures are adapted from the original publication. For full details, please refer to the source article.
By Seungjun Yeo, CEO at Dalton Bioanalytics. Specializing in multi-omics mass spectrometry for drug discovery and biomarker research.