Newswise — Researchers from UNSW Sydney in conjunction with colleagues from Boston University have created a device that exhibits initial potential in identifying Parkinson's disease several years prior to the onset of the initial symptoms.

The scientists detailed in a recent issue of ACS Central Science how they employed neural networks to scrutinize biomarkers in bodily fluids of patients.

The UNSW School of Chemistry scientists investigated blood samples obtained from healthy subjects by the Spanish European Prospective Investigation into Cancer and Nutrition (EPIC). Concentrating on 39 individuals who were diagnosed with Parkinson's disease up to 15 years later, the group employed their machine learning algorithm on datasets containing comprehensive data on metabolites - the chemical substances produced by the body when breaking down food, drugs, or chemicals.

By comparing these metabolites to those of 39 control patients who were matched - individuals in the same study who did not develop Parkinson's - the researchers were able to pinpoint distinct combinations of metabolites that could act as preventive measures or possibly serve as early indicators for Parkinson's disease.

Diana Zhang, a researcher at UNSW, and Associate Professor W. Alexander Donald developed a machine learning software known as CRANK-MS. The acronym stands for Classification and Ranking Analysis using Neural network generates Knowledge from Mass Spectrometry, as explained by Zhang.

“The most common method of analysing metabolomics data is through statistical approaches,” says Ms Zhang.

"To determine which metabolites are more relevant for the disease in comparison to the control groups, scientists typically examine correlations involving certain molecules," explained the researchers.

"However, we take into consideration that metabolites may have connections with other metabolites - and this is where the role of machine learning comes in. With a large number of metabolites ranging from hundreds to thousands, we have utilized computational power to comprehend the underlying mechanisms," added the researchers.

A/Prof. Donald says that in addition to looking at combinations of metabolites, the researchers used an unedited list of data.

"Normally, when researchers utilize machine learning to scrutinize correlations between metabolites and a particular disease, they first reduce the number of chemical features before feeding it into the algorithm," noted the researchers.

"However, in this study, we provide CRANK-MS with all the available data without any feature reduction at the outset. Consequently, we obtain the model's prediction and identify the metabolites that primarily drive the prediction in a single step. This implies that we can now identify metabolites that may have been overlooked using traditional approaches," explained the researchers.

How this could be significant for Parkinson’s Disease

Currently, Parkinson's Disease is diagnosed by observing physical symptoms such as a resting hand tremor, as there are no laboratory or blood tests to diagnose non-genetic cases of the disease. However, atypical symptoms like sleep disorders and apathy can manifest in people with Parkinson's decades before the motor symptoms appear. As a result, CRANK-MS could be utilized when the first signs of these atypical symptoms are detected to determine the likelihood of developing Parkinson's in the future.

Associate Professor Donald stresses that validation studies must be conducted on much larger cohorts, in various regions of the world before the tool can be considered reliable. Nonetheless, the outcomes of the limited cohort evaluated in this study were promising, with CRANK-MS achieving an accuracy of up to 96 percent in analyzing blood chemicals for detecting Parkinson's disease.

"This study is interesting at multiple levels,” he says.

"Firstly, the precision in predicting Parkinson's disease before clinical diagnosis is very high. Secondly, our machine learning technique allowed us to determine the key chemical markers that accurately forecast the development of Parkinson's disease in the future. Thirdly, certain chemical markers that have the greatest impact on precise prediction have previously been linked to Parkinson's disease in cell-based assays but not in humans," elaborated the researchers.

Food for thought

There were some interesting findings when examining the metabolites of people who went on to develop Parkinson’s in the study.

One notable discovery was that triterpenoids were found in lower concentrations in the blood of those who subsequently developed Parkinson's disease than those who did not. Triterpenoids are recognized for their neuroprotective qualities, regulating oxidative stress, and are frequently found in foods like apples, olives, and tomatoes. A subsequent study could investigate whether consuming these foods might naturally safeguard against Parkinson's disease.

Another finding worth investigating further was the presence of polyfluorinated alkyl substances (PFAS) in individuals who later developed Parkinson's disease, which may be linked to exposure to industrial chemicals.

Associate Professor W. Alexander Donald explains, "We have evidence suggesting that PFAS may play a role, but we require additional characterization data to be certain."

Freely available to all

The CRANK-MS tool is freely available to any researchers interested in utilizing machine learning for disease diagnosis with metabolomics data.

“We’ve built the model in such a way that it’s fit for purpose,” says Ms Zhang.

Dr Diana Zhang, one of the researchers involved in the study, suggests that the application of CRANK-MS to detect Parkinson's disease is just one instance of how artificial intelligence can enhance the way we diagnose and monitor diseases. Moreover, CRANK-MS can be effortlessly extended to other diseases to discover new biomarkers of interest.

“The tool is user-friendly where on average, results can be generated in less than 10 minutes on a conventional laptop.”

Journal Link: ACS Central Science