Machine learning helps define new subtypes of Parkinson's disease

Researchers at Weill Cornell Medicine used machine learning to define three subtypes of Parkinson's disease based on the rate at which the disease progresses. In addition to potentially being important diagnostic and prognostic tools, these subtypes are characterized by distinct driver genes. If validated, these markers could also suggest ways to target the subtypes with new and existing drugs.

The study was published July 9. npj digital medicine.

“Parkinson's disease is highly variable, and even people with the same disease can experience very different symptoms,” said lead author Fei Wang, PhD, professor of population health sciences in the Department of Population Health Sciences at Weill Cornell Medical College and founding director of the AI ​​Institute for Digital Health (AIDH). “This indicates that there is probably no one-size-fits-all approach to treating this disease. We may need to consider customized treatment strategies based on a patient's disease subtype.”

The researchers defined subtypes based on distinct patterns of disease progression: They named those with mild baseline disease severity and a slow rate of progression the “slowly progressive” subtype (PD-I, approximately 36% of patients), those with mild baseline disease severity but a moderate rate of progression the “moderate rate of progression” subtype (PD-M, approximately 51% of patients), and those with the fastest rate of symptom progression the “rapid rate of progression” subtype (PD-R).

The research team successfully identified the subtypes by analyzing anonymized clinical records from two large databases using a deep learning-based approach. They also investigated the molecular mechanisms associated with each subtype by analyzing patients' genetic and transcriptomic profiles using a network-based approach. For example, PD-R subtypes activated specific pathways related to neuroinflammation, oxidative stress, and metabolism. The research team also discovered brain imaging and cerebrospinal fluid biomarkers specific to the three subtypes.

Dr. Wang's lab has been studying Parkinson's disease since 2016, when they participated in the Parkinson's Progression Markers Initiative (PPMI) Data Challenge. The team won the challenge focused on subtype derivation and has continued this work ever since. They took data collected from the PPMI cohort as the study's primary subtype development cohort and validated it with the National Institute of Neurological Disorders and Stroke (NINDS) Parkinson's Disease Biomarker Program (PDBP) cohort.

The researchers used their findings to identify potential drug candidates that could be repurposed to target specific molecular changes found in different subtypes, and then used two large, real-world databases of patient health records to confirm that these drugs could help mitigate the progression of Parkinson's disease.

These databases, the New York-based INSIGHT Clinical Research Network, and the OneFlorida+ Clinical Research Consortium, are both part of the National Patient-Centered Clinical Research Network (PCORnet). INSIGHT is led by Dr. Rainu Kaushal, senior vice dean for clinical research at Weill Cornell Medicine and chair of the Department of Population Health Sciences at Weill Cornell Medicine and NewYork-Presbyterian/Weill Cornell Medical Center.

“By examining these databases, we found that people who take the diabetes drug metformin appear to have improved Parkinson's disease symptoms, particularly those related to cognition and falls, compared with people who don't take the drug,” said first author Chang Su, PhD, an assistant professor of population health sciences at Weill Cornell Medical College and a member of AIDH. This was especially true for patients with the PD-R subtypes that are most likely to develop cognitive impairment in the early stages of Parkinson's disease.

“We hope that our study will encourage other researchers to consider using diverse data sources when conducting studies like ours,” said Dr. Wang, “and will enable translational bioinformatics researchers to further validate our findings both computationally and experimentally.”

Numerous collaborators contributed to the research, including scientists from the Cleveland Clinic, Temple University, University of Florida, University of California, Irvine and University of Texas at Arlington, as well as doctoral students from Cornell Tech's computer science program and the computational biology program at Cornell University's Ithaca campus.