New AI-Powered Map Reveals Two Decades of Research Trends

Researchers have just published a comprehensive map, using artificial intelligence to chart the entire research landscape. And yes, they found some interesting, and slightly concerning, treasure spots. A new study published in Analytical Chemistry by Bifarin et al. provides a bird's-eye view of the metabolomics research field, synthesizing insights from over 80,000 publications spanning from 1998 to early 2024. If you've ever tried to keep up with the explosive growth of scientific literature, you'll appreciate the Herculean task of making sense of it all. As the authors note, with research topics diversifying from plant metabolism to human disease, it's increasingly difficult for anyone to track the field's broadening scope.

Hagen Gegner

Scientific Communications Specialist

Visit Bio

Figure 2. Visualization of metabolomics research fields using t-SNE embeddings. (A) Annotated t-SNE projections highlighting research domains.

Charting the Course with AI

So, how did the researchers create this "map of metabolomics"? They employed a sophisticated, domain-specific Large Language Model called PubMedBERT to read and process the abstracts of all 80,656 papers. This process converted the nuanced language of each abstract into a 768-dimensional mathematical representation, capturing its thematic essence.

Using a dimensionality reduction technique known as t-SNE, they projected this complex data onto a two-dimensional map, where publications with similar themes cluster together. The result is a stunning visualization of the entire field, revealing distinct "continents" of research like Plant Biology, Cancer Biology, Pharmacology, and Analytical Chemistry.

The team didn't stop there. To add more detail and clarity, they used GPT-4o mini to refine the clusters into 20 distinct topics. This allowed them to move beyond broad fields and identify specific research hotspots, such as "Plant Stress Response Mechanisms," "Metabolomics in Neurodegenerative Disorders," and the very timely "COVID-19 Metabolomic and Immune Responses".

Putting it to the test, we searched for the keyword "PFAS" (Per- and polyfluoroalkyl substances), the notorious "forever chemicals." The result shows a striking upward trend. Mentions were almost nonexistent before 2010 and then skyrocketed in the last few years

Key Discoveries from the Map

This unique map uncovers several fascinating trends and insights:

The Analytical Chemistry Anchor: Analytical chemistry publications are not only the most numerous but are also spread widely across the map, highlighting the foundational role of analytical methods in all areas of metabolomics. One dense cluster, described as the "horn" of the map, was later decomposed into topics like "Metabolomics Data Analysis and Integration" and "NMR Spectroscopy Innovations," pinpointing where method development is concentrated.

Tracking Emerging Contaminants: The power of the interactive app is that any researcher can explore their own topics of interest. Putting it to the test, we searched for the keyword "PFAS" (Per- and polyfluoroalkyl substances), the notorious "forever chemicals." The result shows a striking upward trend. Mentions were almost nonexistent before 2010 and increased in the last few years, demonstrating how the tool can be used to track the metabolomics community's response to emerging environmental health issues.

The Rise of the Machines: While classical methods like chemometrics are well-represented across the decades, mentions of "deep learning" and "neural networks" show a distinct surge after 2015, clustering in data-science-oriented regions of the map.

The Sample Size Dilemma: The study brings a critical challenge to light: the prevalence of small sample sizes in metabolomics research. An analysis of abstracts revealed that most studies use fewer than 50 samples. While this may be appropriate for methodological work, the authors note it is a recognized bottleneck for biomarker-discovery papers, where underpowered designs can hinder the translation of promising findings into clinical applications.

Watch our Metabolomics Webinar Explore Trends

Sample sizes across all abstracts relevant

The Path Forward

The insights from this manuscript resonate deeply with the push towards automation in the laboratory.

The map reveals a clear trajectory: As metabolomics tackles more complex questions and generates massive datasets for deep learning models, the need for robust, reproducible, and scalable research becomes critical.

This is where automation becomes not just a convenience, but a necessity. The paper highlights the challenge of small sample sizes and the scarcity of clinically validated biomarkers, often due to a lack of validation across independent cohorts. Addressing this requires a significant increase in scale and comparabality. Automated sample preparation platforms, such as the PAL System, are critical for achieving the high throughput needed to run the large, multi-institutional studies the authors call for. This is in line with other researchers also highlighting the problem of reproucibility of biomarkers with at least 85% of proposed biomarkers relating to statistical noise.

Furthermore, the integrity of the data feeding into the "Metabolomics Data Analysis and Integration" and "deep learning" clusters is essential. Automation minimizes human error and analytical variability, ensuring that the data is of the highest quality and consistency. Reducing the variation and noise in the data is one of the most important aspects here.

As the authors state, the future of metabolomics will be defined by a "synergistic relationship between technological advancement... and rigorous biological validation". By enabling larger, more standardized, and more reproducible studies, automation serves as the engine to drive that synergy, helping the field realize its profound potential in improving health and environmental outcomes.

To explore the interactive map yourself, visit https://metascape.streamlit.app/.

For a deeper dive, read the full publication:

Bifarin, O. O., Yelluru, V. S., Simhadri, A., & Fernández, F. M. (2025). A Large Language Model-Powered Map of Metabolomics Research. Analytical Chemistry, 97, 14088-14096. A Large Language Model–Powered Map of Metabolomics Research | Analytical Chemistry

Read Publication