Scientists from University College London and the University Medical Center Goettingen in Germany have developed an AI-enhanced blood test that can predict Parkinson’s disease as far as seven years ahead of symptoms surfacing. This could pave the way for treatments able to slow the progression of the disease, with early detection being a key benefit.
Parkinson’s disease is increasingly a global health concern, affecting almost 10 million people worldwide. Symptoms, caused by the death of nerve cells in the substantia nigra part of the brain, include tremors, movement difficulties, muscle stiffness, balance issues, memory problems, dizziness, and nerve pain. Currently, there are no treatments to halt or reverse disease progression, with diagnosis usually only happening after symptoms have already manifested.
The study started by analyzing blood samples from newly diagnosed Parkinson’s patients and healthy controls. Using advanced mass spectrometry techniques, researchers were able to identify 47 proteins that had different levels of expression between the two groups. They used this to create a blood test that measured the levels of 121 specific proteins. When applied to samples from Parkinson’s patients, healthy individuals, those with other neurological disorders, and patients with isolated REM sleep disorder (a known risk factor for the disease), the test confirmed that 23 proteins showed significant differences between Parkinson’s and healthy subjects.
The breakthrough was the team’s application of machine learning to the data. Models trained to distinguish between Parkinson’s and healthy samples, using only eight proteins, had a 100% accuracy rate in identifying Parkinson’s patients. The test also predicted that 79% of isolated REM sleep disorder samples were like that of someone with Parkinson’s, indicating its potential utility in identifying individuals at high risk of developing the disease over time.
The researchers applied their model to an additional group of 54 isolated REM sleep disorder patients, who provided 146 blood samples. The machine learning model predicted that between 70 and 79% of these samples bore similarities to those of Parkinson’s patients.
Despite requiring larger trials to verify its accuracy and dependability, this AI-powered blood test marks a significant leap in the early diagnosis of Parkinson’s disease. The team anticipates that the process can be simplified further to allow patients to mail a single drop of blood for analysis.
This development builds upon earlier AI-driven efforts to identify and diagnose Parkinson’s disease more efficiently, with other teams previously developing an eye test for early identification. AI continues to be instrumental in combating diseases and accelerating the development of treatments, with notable contributions from projects like Google’s AlphaFold and collaborations between OpenAI and Color Health.
