Parkinson’s Disease May Be Five Distinct Conditions, Machine Learning Study Reveals
A study published in Nature Communications by researchers from VIB and KU Leuven has used machine learning to analyze fruit fly models of Parkinson's disease. The analysis identified two broad subgroups and five smaller subtypes of the condition, suggesting that Parkinson's disease may comprise distinct molecular forms that respond differently to treatments.
Methodology and Findings
The research team created fruit fly models carrying mutations in 24 genes known to be associated with Parkinson's disease. Using unsupervised machine learning, they analyzed behavioral data collected from these models over time without pre-existing hypotheses about the outcomes.
The analysis revealed that the genetic forms of parkinsonism clustered naturally into two main subgroups, which could be further divided into five smaller groups.
"We came in without any preconceived notions of how a specific mutation would affect our animal model." — Dr. Natalie Kaempf, study first author
Subgroup-Specific Drug Responses
When testing potential therapeutic compounds in the animal models, the researchers observed subgroup-specific effects. According to the study, a compound that reversed Parkinson's-related symptoms in one subgroup did not show the same effect in another subgroup.
"We discovered two broad subgroups that can be divided into five smaller groups of parkinsonism." — Patrik Verstreken, VIB-KU Leuven Center for Neuroscience
At the molecular level, the conditions fall into subcategories, and a single drug targeting the different molecular dysfunctions in all cases of Parkinson's disease "essentially doesn't exist."
Potential Applications
The study's findings may support the development of biomarkers and therapies tailored to specific disease subgroups. The researchers noted that the unbiased machine learning approach used in this study could potentially be applied to other diseases that have multiple genetic causes.