Machine learning reveals three distinct subtypes of Parkinson’s Disease
Researchers at Weill Cornell Medicine have utilised advanced machine learning techniques to identify three distinct subtypes of Parkinson’s disease, potentially paving the way for more personalised treatment approaches.
New research from Weill Cornell Medicine has shed light on the heterogeneous nature of Parkinson’s disease, revealing three distinct subtypes characterised by their progression rates. This groundbreaking study, which employed machine learning algorithms to analyse extensive clinical data, could improve the diagnosis, prognosis, and treatment of this complex neurodegenerative disorder.
The study, published on 10 July 2024 in the journal npj Digital Medicine [1], represents a significant step forward in our understanding of Parkinson’s disease and its varied manifestations.
Unveiling the subtypes
The research team, led by Dr Fei Wang, professor of population health sciences and founding director of the Institute of AI for Digital Health at Weill Cornell Medicine, utilised deep learning-based approaches to scrutinise deidentified clinical records from two large databases. Their analysis revealed three distinct subtypes of Parkinson’s disease, each characterised by a unique progression pattern:
- Inching Pace (PD-I): Comprising approximately 36% of patients, this subtype exhibits mild baseline severity and slow progression.
- Moderate Pace (PD-M): Accounting for about 51% of patients, this subtype is characterised by mild baseline severity but advances at a moderate rate.
- Rapid Pace (PD-R): This subtype demonstrates the most rapid symptom progression rate.
Dr Wang emphasised the importance of these findings: “Parkinson’s disease is highly heterogeneous, which means that people with the same disease can have very different symptoms. This indicates there is not likely to be a one-size-fits-all approach to treating it. We may need to consider customised treatment strategies based on a patient’s disease subtype.”
Molecular mechanisms and biomarkers
Beyond identifying the subtypes, the researchers delved deeper into the molecular mechanisms associated with each. Through the analysis of patient genetic and transcriptomic profiles using network-based methods, they uncovered distinct pathways and biomarkers for each subtype.
For instance, the PD-R subtype demonstrated activation of specific pathways related to neuroinflammation, oxidative stress, and metabolism. The team also identified distinct brain imaging and cerebrospinal fluid biomarkers for each of the three subtypes.
Repurposing existing drugs
One of the most promising aspects of this research is its potential to inform new treatment strategies. By leveraging their findings, the researchers identified possible drug candidates that could be repurposed to target the specific molecular changes observed in the different subtypes.
To validate these potential treatments, the team analysed two large-scale, real-world databases of patient health records: the INSIGHT Clinical Research Network and the OneFlorida+ Clinical Research Consortium. Both are part of the National Patient-Centered Clinical Research Network (PCORnet).
Dr Chang Su, first author of the study and assistant professor of population health sciences at Weill Cornell Medicine, reported an intriguing finding: “By examining these databases, we found that people taking the diabetes drug metformin appeared to have improved disease symptoms – especially symptoms related to cognition and falls – compared with those who did not take metformin.”
This effect was particularly pronounced in patients with the PD-R subtype, who are most likely to experience cognitive deficits early in the course of their Parkinson’s disease.
The power of diverse data sources
The study’s innovative approach, combining machine learning techniques with diverse data sources, has the potential to advance Parkinson’s disease research. Dr Wang expressed hope that their work would inspire other investigators to consider using diverse data sources in similar studies.
He added: “We also think that translational bioinformatics investigators will be able to further validate our findings, both computationally and experimentally.”
While these findings represent a significant advancement in our understanding of Parkinson’s disease, further research is needed to validate and expand upon these results. If confirmed, this subtype classification system could become an invaluable diagnostic and prognostic tool, enabling clinicians to tailor treatment strategies to individual patients based on their specific subtype.
Moreover, the identification of distinct driver genes and molecular pathways for each subtype opens up new avenues for drug development and repurposing, potentially leading to more effective treatments for this debilitating condition.
Reference:
Su, C., et al. (2024). Identification of Parkinson’s disease PACE subtypes and repurposing treatments through integrative analyses of multimodal data. npj Digital Medicine, 7, 98. https://doi.org/10.1038/s41746-024-00789-9
