Scientists from the Quadram Institute and the European Molecular Biology Laboratory (EMBL) combined almost 4,500 samples from 22 different studies carried out across the world and then used machine learning techniques to analyse this vast data set.

As well as identifying the types of bacteria most associated with Parkinson’s disease, this meta-analysis also highlighted the metabolic functions most commonly seen in Parkinson’s disease gut microbiomes that could contribute to poor gut health. Intriguingly, microbial pathways potentially involved in biochemical transformation of solvents and pesticides are enriched in Parkinson’s disease, tying in with other studies that linked exposure of such chemicals with the condition.

Parkinson’s disease is an increasingly common neurodegenerative condition, characterised by involuntary shaking, stiffness and slow movement. It also causes many other symptoms not related to movement, including gut problems like constipation, inflammation and leakiness of the gut lining. In some cases, the gut disorders appear decades before the motor symptoms. This has led to interest in the role of the gut microbiome in understanding, diagnosing and even potentially treating the disease.

Several studies analysing the microbiome of people with Parkinson’s have shown differences from a healthy microbiome, but there is a lot of variation between these studies. This isn’t unusual in microbiome studies, where there is often variability in the methodology used, but also inherent variations in populations around the world.

An earlier meta-analysis by Quadram Institute researchers  published four years ago, combining data from different studies, identified some common features of the Parkinson’s microbiome across different continents, but to date a consensus on the microbes that characterise the Parkinson’s microbiome is missing.

Identifying the microbes, and metabolic functions, associated with Parkinson’s disease is essential to unpicking the mechanisms that might potentially link the microbiome to the condition, and would open the door to future research into microbiome-based strategies for diagnosis and treatment.

Because of the vast amount of data associated with analysing the trillions of microbes that make up the gut microbiome, researchers have turned to machine learning techniques to look for patterns that could discern the Parkinson’s.

Machine learning is related to Artificial Intelligence in that it uses statistical principles to identify and learn patterns in large data sets, and to make predictions on future data points.

Some individual studies have shown promise, reporting 90% accuracy in their models, but the real test comes when the prediction models are used on samples from a different, independent population.

Researchers from the Quadram Institute joined colleagues from the European Molecular Biology Laboratory in Heidelberg, Germany and carried out a large-scale meta-analysis to construct and evaluate such machine learning models in a unified manner on a large pooled dataset of publicly available microbiome data from 22 studies spanning four continents.

Their finding, published in the journal Nature Communications, suggested that most of the machine learning models were study specific and did not generalise well to other studies. This may have been because most of the studies were limited to small samples of people with Parkinson’s that weren’t representative of the larger global patient population, and didn’t encompass variabilities in this complex condition. However, by training machine learning models on data from many studies at the same time, models became considerably better in recognising Parkinson’s-related microbiome changes in other studies as well.

A striking finding from the team’s new analysis was that Parkinson’s disease microbiomes were enriched in genes involved in breaking down xenobiotics, which are chemicals that are foreign to the body, and include pesticides, solvents and pollutants.

It’s not clear which xenobiotics have induced this signature in the microbiome, and it can’t be ruled out that Parkinson’s medication could generate the effect seen. Nor is it clear what effect an enriched ability by the microbiome to break them down has.

“It is intriguing to speculate that the composition of the gut microbiome might be altered as a consequence of exposure to these chemicals. Or could the breakdown of these chemicals by the microbiome change their known effects on the neurons in the brain? And could this be a protective effect, or could the process increase the neurotoxicity of these chemicals?” commented Dr Georg Zeller, senior author of the study, based at EMBL.

“More research is needed to understand the molecular mechanisms of how gut microbes transform these chemicals.”

The researchers’ meta-analysis also found hallmarks of pathogenic bacteria involved in infection-like processes that may contribute to inflammation and increased gut permeability.

A leaky gut could facilitate the translocation of bacterial products and potentially toxic compounds from the gut into the body and potentially into the brain and central nervous system.

More research is warranted to study and verify this, but as our gut microbiome is individual to each of us, the implication is our microbiomes could increase our risk of developing Parkinson’s Disease or even provide some detoxification-based protection against it.

“By leveraging large amounts of data we’ve provided the most up to date view on the taxonomic and functional features of the gut microbiome robustly associated with Parkinson’s disease,” commented first author Dr Stefano Romano, first author on the study who worked at both the EMBL and the Quadram Institute.

Professor Arjan Narbad from the Quadram Institute commented “There is increasing evidence on the role of gut microbiome in aetiology of Parkinson’s disease but variation in individual microbiomes makes it difficult to pinpoint the specific microbes most likely to be involved in disease processes. Combining the use of machine learning with in-depth metagenomic sequencing using large cohorts provides a promising basis for diagnosis and therapeutic potential.”

The research was supported by the Biotechnology and Biological Sciences Research Council (BBSRC), part of UKRI, EMBL, LUMC, the Federal Ministry of Education and the Deutsche Forschungsgemeinschaft.

Reference: Machine learning-based meta-analysis reveals gut microbiome alterations associated with Parkinson’s disease, Stefano Romano, Jakob Wirbel, Rebecca Ansorge, Christian Schudoma, Quinten Raymond Ducarmon, Arjan Narbad & Georg Zeller, Nature Communications 16, 4227 (2025). DOI: 10.1038/s41467-025-56829-3

About EMBL: The European Molecular Biology Laboratory (EMBL) is Europe’s life sciences laboratory. We provide leadership and coordination for the life sciences across Europe, and our world-class fundamental research seeks collaborative and interdisciplinary solutions for some of society’s biggest challenges. We provide training for students and scientists, drive the development of new technology and methods in the life sciences, and offer state-of-the-art research infrastructure for a wide range of experimental and data services.

EMBL is an intergovernmental organisation with 29 member states, one associate member, and one prospective member. At our six sites in Barcelona, Grenoble, Hamburg, Heidelberg, Hinxton near Cambridge, and Rome, we seek to better understand life in its natural context, from molecules to ecosystems.