Large Language Model-Generated Dietary Metabolite Biomarker Database Drives Deep Annotation of the Human Diet Metabolome

Clinical Snapshot

Key Findings

• P Value: Not reported

• Effect Size: 3.7-fold increase in annotation yield (from 566 to 2,078 metabolites) with dual-strategy versus LC-MS database alone; DMBDB contains 4,983 non-redundant dietary biomarkers derived from 7,339 articles

• Primary Outcome: Annotation coverage of dietary biomarkers in human urine via untargeted LC-HRMS: 566 metabolites annotated by the LC-MS database alone; 2,078 total annotations achieved with integration of SGMNS — a 3.7-fold increase in coverage

• Nnt Or Sensitivity: F1 score for biomarker name recognition by LLM text-mining pipeline: 0.9269 (precision and recall components not separately reported in abstract)

• Confidence Interval: Not reported

Clinical Application

Implementation requires access to LC-HRMS instrumentation, bioinformatics infrastructure capable of running molecular networking (e.g., GNPS/FBMN), and the DMBDB itself (availability not confirmed at time of publication). The computational pipeline for database construction is described as automated but would require significant technical expertise to deploy or adapt. Routine clinical laboratory adoption is not currently feasible without further validation and commercial or open-access tool development. Australian nutritional epidemiology research (e.g., through CSIRO, the Australian Longitudinal Study on Women's Health, or the Australian Dietary Biomarker Initiative) could benefit from an expanded dietary metabolite database. However, the DMBDB's coverage of Australian-specific dietary patterns — including Indigenous Australian foods, native botanicals, and the Australian dietary supplement market (TGA-regulated) — is unknown and likely limited. The PBS does not currently fund metabolomic dietary biomarker testing, and the TGA has not evaluated this tool. RACGP guidelines on dietary assessment rely on validated dietary recall instruments rather than metabolomics; this technology remains a research tool. Australian researchers using platforms such as Metabolomics Australia would need to validate DMBDB performance against locally relevant food composition databases (e.g., AUSNUT). Researchers and clinical scientists conducting untargeted metabolomics in nutritional epidemiology, dietary exposure assessment, diet-disease association studies, and food biomarker discovery. Potentially applicable to clinical settings where dietary compliance monitoring via urinary metabolomics is of interest (e.g., dietary intervention trials, cardiometabolic disease research).

Abstract

The annotation of dietary biomarkers is crucial for nutritional epidemiology. While untargeted liquid chromatography-high-resolution mass spectrometry (LC-HRMS) is a powerful analytical approach, the annotation of dietary biomarkers is hampered by the low specificity of existing public databases, which limits annotation coverage and accuracy. To address this limitation, we developed a novel database construction strategy and a dual-annotation workflow. We first employed an automated, large language model (LLM)-based text-mining pipeline to parse 7339 scientific articles and supplementary materials, creating the Dietary Metabolite Biomarker Database (DMBDB), which contains 4983 nonredundant biomarkers. The LLMs workflow demonstrated high performance, achieving an F1 score of 0.9269 for biomarker name recognition. Subsequently, two complementary annotation strategies were designed: (i) a specialized LC-MS database derived from DMBDB, incorporating predicted retention times and experimental MS/MS spectra for high-confidence matching, and (ii) a structure-guided molecular networking strategy (SGMNS) that uses DMBDB as background knowledge to annotate dietary biomarkers and their metabolites lacking spectral evidence. The framework was validated using untargeted LC-HRMS analysis of urine samples. LC-MS database directly annotated 566 metabolites, and the integration with SGMNS expanded the total number of annotations to 2078. The LLM-driven database construction combined with the dual-strategy annotation framework provides a powerful paradigm for achieving high-coverage and high-accuracy dietary metabolomics.

References

1. 1.Nie, Z., Zheng, F., Hu, D., Fu, Z., & Cao, C. (2026). Large language model-generated dietary metabolite biomarker database drives deep annotation of the human diet metabolome. Analytical Chemistry. Advance online publication. https://doi.org/10.1021/acs.analchem.6c01612