AsianScientist (Jun. 25, 2026) – Cancer has a way of rewiring the body’s metabolism to fuel tumour growth. As metabolites are the chemical products left behind by these cellular processes, studying them can reveal telltale signs of disease. This makes metabolomics particularly helpful for pancreatic cancer, where early detection is still exceedingly difficult.
The blood biomarker CA19-9, currently used to diagnose and monitor pancreatic cancer, is far from perfect. Its levels can rise in other cancers and even benign conditions. Although previous studies have explored alternative metabolic markers, none have consistently outperformed CA19-9 in clinical practice.
In a new study published in Nature Communications, a multidisciplinary team from National Taiwan University Hospital and Academia Sinica developed a diagnostic tool called PanMETAI. By combining advanced metabolomics with artificial intelligence, the platform pieces together patterns in the patient’s blood chemistry, biomarker levels and clinical information to spot pancreatic cancer with high accuracy.
The researchers used high-resolution proton nuclear magnetic resonance (1H NMR) spectroscopy, a technique that can generate detailed chemical fingerprints of blood serum without complex sample preparation. Unlike many previous mass spectrometry-based metabolomics approaches that focused on only a handful of molecules, the method captures a broad snapshot of the extensive metabolic changes that drive cancer.
These metabolic signatures were combined with other clinically relevant information, including patient age, levels of the standard cancer marker CA19-9 and a protein called Activin A, which has been linked to pancreatic cancer progression.
To navigate this complex data, the researchers vetted several state-of-the-art AI systems. They ultimately selected a TabPFN model, a type of AI designed to detect subtle relationships across large numbers of variables, to power PanMETAI.
Using data from 902 participants in Taiwanese cohorts, PanMETAI achieved near-perfect discrimination between cancer patients and high-risk controls, including those with early-stage disease. Importantly, the model maintained its strong performance when validated in an independent Lithuanian cohort of 322 participants. This demonstrated its robust algorithm for detecting pancreatic cancer across diverse populations.
The researchers were particularly intrigued that PanMETAI honed in on a relatively small set of meaningful metabolic changes that are well-known hallmarks of cancer. This included lower HDL cholesterol levels, higher glucose and lactate levels and disruptions in amino acid metabolism.
Beyond its accuracy, PanMETAI offers several practical advantages that could support its implementation in hospitals and research centres. The platform is built on a standardised NMR-based workflow that produces consistent results. It also performed well even when trained on small patient cohorts, reducing the need for large datasets which could be hard to obtain. In addition, the predictions remained stable even with potential confounding factors and across samples collected over nearly two decades.
“By combining the power of AI with the rich metabolic information captured by NMR spectroscopy, we have created a tool that can detect pancreatic cancer at its earliest and most treatable stages,” said study author Yu-Ting Chang, a professor of internal medicine at National Taiwan University.
“Our goal is to bring this technology to clinical practice so that more patients can benefit from timely diagnosis and treatment,” he added.
—
Source: National Taiwan University Hospital ; Image: TinaJi/Magnific
This article can be found at PanMETAI – a high performance tabular foundation model for accurate pancreatic cancer diagnosis via NMR metabolomics.
Disclaimer: This article does not necessarily reflect the views of AsianScientist or its staff.
