A new study led by researchers at VIB and KU Leuven shows that Parkinson’s disease can be divided into different subtypes, helping to explain why no single treatment is effective for all patients. Using machine learning analysis, the team identified two main groups and five subgroups of the disease, marking an important step toward more personalized treatment. The results of this research have recently nature communications.
“We have discovered two major subgroups that can be divided into five smaller groups of parkinsonism,” says Professor Patrick Verstreken (VIB-KU Leuven Neuroscience Center).
Parkinson’s disease affects millions of people worldwide and is traditionally defined by clinical symptoms such as movement difficulties and progressive neurological decline. However, despite being classified as a single disease, Parkinson’s disease can be caused by mutations in many different genes, and the underlying biological mechanisms are diverse. This complexity poses challenges to the development of effective treatments, as treatments targeting one pathway may not be effective in all patients.
New research reveals that these genetically distinct forms of Parkinson’s disease can be classified into distinct molecular subtypes, highlighting the need to rethink the disease as a collection of related symptoms and open the door to more targeted therapeutic approaches.
“When clinicians and patients look at this disease, they see the clinical symptoms, and that’s what people associate with Parkinson’s disease,” says Verstrieken. “But if you look under the hood at the molecular level, you find that they fall into subcategories. And this is important because there is essentially no single drug that targets the different molecular dysfunctions in all Parkinson’s diseases.”
Rather than starting with assumptions about how different genetic mutations affect the disease, the researchers monitored the behavior of fruit fly models with mutations in Parkinson’s disease-related genes over time and used unbiased computational and machine learning-based methods to identify patterns. By letting the data guide their analysis, the team was able to uncover natural groupings of diseases in these animals that traditional hypothesis-based methods could not reveal.
“We went into the study without any preconceptions about how a particular mutation would affect an animal model. We simply took animals with mutations in one of 24 different disease-causing genes and monitored their behavior over time,” added Dr. Natalie Kemp, lead author of the study.
Taken together, this unbiased strategy reveals previously hidden structures within Parkinson’s disease, showing that different genotypes naturally cluster into distinct subtypes.
By moving away from assumptions and allowing patterns to emerge directly from the data, this study provided a powerful framework for understanding the biological diversity of disease and guiding future research toward more precise interventions. This is also an excellent example of how machine learning can uncover otherwise undetectable features of disease biology, revealing hidden structures and clinically meaningful variations not apparent with traditional approaches.
“We know that there are different types of Parkinson’s disease,” says Verstrieken. “Having these subcategories allows us to look at groups of patients with specific mutations, search for specific biomarkers, and develop drugs tailored to each group.”
Researchers were able to cure the Parkinson’s disease phenotype in animal models by testing compounds in different subgroups. They also observed that different subgroups responded differently to different compounds.
“When we took the first compound that cured subgroup A and tested it on subgroup B, it didn’t save the latter. Our study shows that we can create subgroup-specific drugs that have a positive effect and are actually specific for that subgroup,” Verstrieken explains.
And this unbiased strategy could be adopted for other diseases caused by mutations in multiple genes.
“The same principle can be applied to other types of diseases. Diseases caused by mutations in a variety of different genes or environmental factors could potentially be classified according to this principle,” concludes Verstreken.
sauce:
Flemish Institute of Biotechnology
Reference magazines:
Kemp, N. others. (2026). Behavioral screening defines molecular parkinsonism-associated subgroups in Drosophila. Nature Communications. DOI: 10.1038/s41467-026-70303-8. https://www.nature.com/articles/s41467-026-70303-8
