Artificial intelligence has been used to successfully predict the onset of Parkinson’s disease nearly 15 years before symptoms appear, through research conducted by a team of global experts.
Scientists from UNSW Sydney and Boston University have collaborated to develop a machine learning program that can analyze biomarkers in blood samples for metabolites typically found in people with Parkinson’s disease.
Unique Patterns Identified Exposure to industrial chemicals has been identified as a potential cause of Parkinson’s. A neuroprotective metabolite found in apples, olives and tomatoes has been found at higher levels in people who don’t have Parkinson’s disease. Future research might better understand the risk factors or risk reducers
It could be a major breakthrough for people with Parkinson’s, since no blood or laboratory test is available to diagnose non-genetic cases. Currently, people have to rely on the diagnosis once physical symptoms such as hand tremor at rest appear. However, a successful blood test could diagnose the disease much earlier.
It is estimated that over 100,000 Australians have Parkinson’s disease – some figures put the number at 219,000 people – and around 32 new cases are diagnosed every day.
As part of the research, the scientists were able to compare data collected up to 15 years before an official diagnosis of Parkinson’s. With a vast amount of information, they could take a new approach to analyzing blood samples.
How does it work? Metabolites are formed during metabolism, the process by which food, chemicals, and tissues are converted into energy and materials needed for growth and repair. Scientists studied 39 patients who developed Parkinson’s and 39 who didn’t to identify unique combinations of metabolites that could act as early warning signs. They developed a machine learning tool that could detect Parkinson’s up to 15 years before symptoms appear realized
“The most common method for analyzing metabolomics data is statistical approaches,” explained researcher Diana Zhang. “To find out which metabolites are more important for the disease compared to controls, researchers typically look at correlations involving specific molecules.”
“But here we take into account that metabolites can have associations with other metabolites – and this is where machine learning comes into play. With hundreds to thousands of metabolites, we used computational power to understand what was going on.”
Artificial intelligence provided the computing power for CRANK-MS – also known as the rather clunky classification and ranking analysis using neural networks that generates knowledge from mass spectrometry.
For CRANK-MS there was no need to filter information as hundreds of different combinations of metabolites could be analysed. It would be nearly impossible for humans to analyze these combinations unaided.
Despite a 96% success rate in detecting Parkinson’s disease, Associate Professor W. Alexander Donald acknowledged that the high accuracy rate is not common in clinical diagnosis. He said that due to the small sample size, more research needed to be done.