Dog noses and AI provide new clues about Long COVID

What dogs can sniff out can be revealed by mass spectrometry: researchers from Technische Universität Braunschweig, Hannover Medical School and the University of Veterinary Medicine Hannover have shown that post-COVID syndrome can be identified on the basis of odour signatures. The results also reveal a correlation between the assessments of specially trained sniffer dogs and modern mass spectrometric analyses combined with machine learning techniques. The researchers are thus providing new insights into disease-specific metabolic patterns and opening up prospects for innovative diagnostic approaches.

Why can specially trained dogs detect people with Long COVID? And can the odour signatures perceived by the animals also be detected using modern analytical methods and artificial intelligence? These are the questions addressed by the ‘COVID Dogolomics’ research project, on which researchers from Technische Universität Braunschweig, Hannover Medical School (MHH) and the Hannover University of Veterinary Medicine (TiHo) are collaborating. The research was recognised at the closing symposium of the Lower Saxony COVID-19 Research Network (COFONI). The findings will also be presented at the international conference ‘Metabolomics 2026’ in Buenos Aires.

In search of objective markers for Long COVID

Although millions of people worldwide are affected by Long COVID, objective diagnostic methods are still lacking. Symptoms such as chronic fatigue, difficulty concentrating, breathing difficulties or exercise intolerance also occur in other conditions, making a clear diagnosis difficult.

“For many of those affected, the situation remains difficult to this day because we still do not fully understand the biological processes underlying Long COVID,” says Professor Karsten Hiller, Head of the Department of Bioinformatics and Biochemistry at TU Braunschweig. “That is why we are looking for measurable metabolic changes that can help us to better characterise the condition and, in the long term, develop more objective diagnostic methods.”

The analytical basis for the project was established at TU Braunschweig. In her PhD thesis, Lea Woyciechowski developed a new method for analysing volatile metabolites in very small urine samples. The methodology, published in the journal ‘Metabolites’, forms the basis for the studies now being presented. Using this method, so-called volatile organic compounds (VOCs) can be detected at high resolution and utilised for further analysis using machine learning techniques.

The project brings together the clinical expertise of Hannover Medical School under Prof. Dr. Georg Behrens, research into medical detection dogs at the University of Veterinary Medicine Hannover led by Prof. Dr. Holger Volk, and the analytical and bioinformatic investigations of the Department of Bioinformatics and Biochemistry at TU Braunschweig, led by Prof. Dr. Karsten Hiller. Together, the partners aim to gain a better understanding of the biological signatures of post-COVID syndrome and, in the long term, to develop new diagnostic approaches.

As part of the project, Hannover Medical School provided patient cohorts and biological samples. Hannover University of Veterinary Medicine used specially trained sniffer dogs to investigate whether Long COVID samples could be identified by their smell. Researchers at TU Braunschweig analysed the same samples using state-of-the-art mass spectrometry and developed machine learning methods to decipher the underlying metabolic patterns.

The results

The results were remarkable: the dogs were able to reliably distinguish Long COVID samples from healthy control samples and even from similar clinical presentations. This suggests that Long COVID may be associated with a characteristic odour signature. But what biological changes lie behind this odour?

From the dog’s nose to mass spectrometry

This is where Lea Woyciechowski’s work comes in. The PhD student in the Department of Bioinformatics and Biochemistry has developed a new analytical method that enables high-resolution detection of volatile organic compounds (VOCs) from very small urine samples. These molecules are produced as a result of metabolism and can provide clues to physiological or disease-related processes.

“Volatile metabolites are, in a sense, the chemical fingerprints of biological processes,” explains Lea Woyciechowski. “We wanted to find out whether the signature that dogs perceive can also be analytically detected and described using data-driven methods.”

In doing so, characteristic patterns were identified that distinguish Long COVID samples from control groups.

Two completely different systems detect the same signature

It is particularly interesting that the dogs’ findings and the analytical evaluations corresponded surprisingly well: samples identified as abnormal by the dogs also exhibited characteristic metabolic patterns in the statistical models. This means that two fundamentally different systems point to the same disease-associated changes.

“The fact that two completely different detection systems independently recognise the same signature is particularly exciting from a scientific perspective,” says Professor Hiller. “This gives us additional assurance that we are indeed observing relevant biological changes and not merely statistical random effects.”

The results thus provide new evidence that post-COVID syndrome is linked to measurable changes in metabolism. At the same time, they demonstrate the potential of combining biological sensor systems with modern data analysis.

Next steps: The molecules behind Long COVID

The researchers are now facing the next important step. Although several candidates have already been identified that play a key role in distinguishing between the samples, their exact chemical structure has not yet been fully elucidated. Over the coming years, these molecules are to be unambiguously identified and subsequently verified experimentally. The key question is: are these really the compounds that the dogs are detecting?

Partners

The project involves the Department of Bioinformatics and Biochemistry at Technische Universität Braunschweig at the Braunschweig Integrated Centre of Systems Biology (BRICS), the Hannover Medical School and the University of Veterinary Medicine Hannover. The research was carried out as part of the Lower Saxony COVID-19 Research Network (COFONI) and is also embedded within the scientific community of MetaBoSpace, a network for metabolism-oriented research at TU Braunschweig.