New AI Tool Predicts Parkinson's Disease Years Before Symptoms Appear

A new AI tool has been developed by scientists at the University of New South Wales Sydney in collaboration with Boston University, showing promising potential in detecting Parkinson's disease years before the manifestation of symptoms.

By employing neural networks to analyze biomarkers found in patients' bodily fluids, the researchers successfully pinpointed unique combinations of metabolites that could potentially act as early indicators or preventive signals for Parkinson's disease.

CRANK-MS

In their study, the researchers examined blood samples from healthy individuals that were gathered as part of the Spanish European Prospective Investigation into Cancer and Nutrition (EPIC).

The team specifically focused on 39 patients who later developed Parkinson's disease, even up to 15 years after the samples were collected. Utilizing a machine learning program, they meticulously analyzed datasets that contained comprehensive information about metabolites-molecules formed during the breakdown of food, drugs, or chemicals.

By comparing these metabolites with those of 39 control patients who did not develop Parkinson's, the researchers successfully identified distinct combinations of metabolites that were closely associated with the disease.

UNSW researcher Diana Zhang and Associate Professor W. Alexander Donald have introduced CRANK-MS, an innovative machine learning tool.

CRANK-MS stands for Classification and Ranking Analysis using Neural network generates Knowledge from Mass Spectrometry. It surpasses traditional statistical methods in analyzing metabolomics data by considering metabolite associations and interconnectedness.

Unlike conventional methods that involve reducing the number of chemical features, the researchers took a distinct approach with CRANK-MS. They inputted all available information into the tool without data reduction.

This strategy enabled CRANK-MS to generate model predictions and identify the key metabolites in a single step. By adopting this approach, the researchers enhanced the likelihood of capturing previously overlooked metabolites.

How AI Could Improve Parkinson's Detection

Parkinson's disease is currently diagnosed based on physical symptoms, without blood or laboratory tests for non-genetic cases. CRANK-MS can be used when atypical symptoms appear to assess future Parkinson's risk.

Validation studies on larger cohorts are needed, but the limited study showed promising results with CRANK-MS achieving up to 96% accuracy in detecting Parkinson's.

While the study provides intriguing insights, further investigation is needed. For instance, lower concentrations of triterpenoids were observed in the blood of individuals who later developed Parkinson's, whereas polyfluorinated alkyl substances (PFAS) were found in those who developed the disease, potentially indicating exposure to industrial chemicals.

The findings suggest further research opportunities, including investigating the potential protective effects of triterpenoid-rich foods like apples, olives, and tomatoes against Parkinson's disease.

The researchers anticipate that CRANK-MS can be utilized for other diseases to discover new biomarkers of interest, highlighting the transformative potential of AI in disease detection and comprehension.

The study was published in the journal ACS Central Science.