Skip to main content
. 2026 Sep 22;16:1934250. doi: 10.3389/fcimb.2026.1934250

Figure 2.

Infographic in two panels. Panel A traces four classes of candidate signature — taxonomic community profiling, Helicobacter pylori-specific molecular targets, metabolic and volatile markers, and computational integration — from the specimen in which each is measured (saliva, dental plaque, tongue coating, breath) through what is measured and the readout, to three target conditions: gastric H. pylori infection, histological gastritis, and clinically meaningful disease activity. A caveat row notes that most candidates detect microbial presence rather than disease activity. Panel B places the same candidates on a grid of evidence population (pediatric, adult, non-human) against stage of biomarker development, with the longitudinal and clinical-impact row empty in every population.

Specimen-analyte-target mapping and evidence status of candidate signatures. (A) Four candidate signature classes—taxonomic (community), H. pylori-specific molecular, metabolic/volatile, and computational/integrative—are traced from the specimens in which each is actually measured, through what is measured, to the readout, and each class lists only the specimens used for that analyte: 16S rRNA gene sequencing and shotgun metagenomics against saliva, dental plaque and tongue coating; nucleic-acid detection of the H. pylori 23S rRNA gene, cagA and sabA, and of exosomal H. pylori DNA and clarithromycin-resistance markers, against saliva and dental plaque; nitrate-nitrite profiling and breathomics against saliva and breath; and machine-learning models against multi-matrix input. Salivary nucleic-acid targets are not breath-based, so breath appears only against volatile and urea-derived analytes. Reported pediatric accuracy is given where available. The output row carries the three target conditions defined in Section 2, gastric H. pylori infection, histological gastritis including its H. pylori-negative fraction, and clinically meaningful disease activity, each of which must be prespecified together with its reference standard. The final row is a deliberate caveat: most candidates detect microbial presence (colonization/dysbiosis) rather than disease activity, and only host-response-linked signatures address the activity-level endpoint. (B) The same candidates are placed on the two grading axes of Section 2, with the evidence population across the columns (pediatric human, adult human, and non-human or analytical) and the stage of development down the rows. Each row label names the study design and the corresponding phase, so the panel can be read against Tables 2 and 3 without cross-reference. Two pediatric candidates occupy the Phase 2 row, salivary 23S rRNA gene PCR and the salivary exosomal assay, and both address gastric infection rather than gastritis. The top row, covering Phases 3 to 5, is empty in every population: independent-cohort validation has been reported for selected adult gastric-cancer signatures and is recorded within Phase 2, but no candidate in any population has reached the longitudinal, prospective intended-use, or clinical-impact stages, and no pediatric candidate has progressed beyond Phase 2 at all, which is the review’s central gap. Taxa and gene symbols are italicized throughout; protein products are set in roman. Panel (B) carries no tier colors and no direction marks; the evidence population and the development phase are given by the row and column position. GC, gastric cancer; VOC, volatile organic compound; ML, machine learning.