Recently, a research team led by Prof. XU Guowang and Prof. LIU Xinyu from the Dalian Institute of Chemical Physics (DICP) of the Chinese Academy of Sciences (CAS) has proposed a framework for "Third-generation metabolomics" that integrates standardization, streamlined workflow, automation, and intelligence.
The study was published online in TrAC-Trends in Analytical Chemistry on October 10, 2026.
Agent-driven third-generation metabolomics unifies standardization, automation, intelligence, and streamlining to enable closed-loop automation and systems biology integration. (Image by Jing Fengyang)
For decades, mass spectrometry-based metabolomics has largely relied on two main approaches: untargeted and targeted strategies. Untargeted metabolomics can detect a wide range of metabolites but faces challenges in reproducibility and data processing. Targeted metabolomics, by contrast, offers more reliable quantitative results but typically measures fewer metabolites.
To combine the advantages of these approaches, Prof. XU's team introduced pseudotargeted metabolomics in 2012. This hybrid strategy was later described by Prof. Oliver Fiehn as "Second-generation metabolomics" and has since been applied in biomedical, agricultural, and environmental research. However, as metabolomics increasingly supports high-throughput, large-scale, and multicenter studies, challenges remain in ensuring analytical reproducibility, comparability, and effective data mining.
To address these challenges, the researchers proposed a "Third-generation metabolomics" framework built around four interconnected principles: standardization, streamlined workflow, automation, and intelligence. Standardization ensures consistency in sample collection and analysis, while streamlined workflow connects different stages of the analytical process. Automation reduces manual intervention, and AI-driven intelligence enhances data interpretation and knowledge discovery.
Going beyond the automation of individual tasks, the framework envisions AI agents coordinating the entire workflow, from experimental design and instrument operation to data processing and interpretation. Through feedback-driven optimization and adaptive decision-making, these agents could enable more integrated and autonomous metabolomics workflows.
The researchers also outlined two potential directions for the development of "Third-generation metabolomics".
First, integrating automated sample preparation, mass spectrometric analysis, real-time quality control, and intelligent data processing could enable "Lights-out" laboratories, improving analytical efficiency and data quality.
Second, combining high-quality metabolomics data with multi-omics information, knowledge graphs, and AI models, could enable the development of "Metabolome-centered digital twins". These models could simulate metabolic states and track their dynamic changes across different tissues and organs, offering new opportunities to investigate complex biological systems.
Overall, the proposed framework offers a path from fragmented, expert-dependent procedures toward more reproducible and intelligent metabolomics. It could also help move the field beyond static descriptions of metabolic phenotypes toward dynamic modeling and prediction, supporting future research in precision medicine.