A new machine learning tool called CRANK-MS was able to identify with high accuracy those people who could develop Parkinson’s disease based on the analysis of blood molecules.
This algorithm identified several molecules that may serve as early biomarkers for Parkinson’s disease.
According to researchers at Australia’s University of New South Wales (UNSW), who are developing machine learning tools with their colleagues at Boston University in the United States, these findings show the potential for artificial intelligence (AI) to improve healthcare. It is said that there is
“The application of CRANK-MS to Parkinson’s disease detection is just one example of how AI can improve how we diagnose and monitor disease,” UNSW study co-author Diana Zhang said in a press release. said in
the study, “Interpretable machine learning of metabolomics data reveals Parkinson’s biomarkers]was published. ACS Central Science.
Analyze more data with CRANK-MS machine learning tools
Parkinson’s disease is now diagnosed based on the symptoms a person is experiencing. There is no biological test that can definitively identify this disease. Many researchers are working to identify biomarkers for Parkinson’s disease, which may be measured to identify or predict the risk of developing neurodegenerative disease.
Here, an international team of researchers used machine learning to analyze metabolomics data, a large-scale analysis of the levels of thousands of different molecules found in patients’ blood, to identify biomarkers for Parkinson’s disease.
Blood samples collected from the European Prospective Investigation on Cancer and Nutrition (EPIC) in Spain were used for the analysis. There were 39 samples from people who developed Parkinson’s disease after up to 15 years of follow-up, and another 39 samples from people who did not develop Parkinson’s disease throughout the follow-up period. The metabolomics composition of the samples was evaluated using a chemical analysis technique called mass spectrometry.
At its simplest, machine learning involves feeding computers large amounts of data along with a set of goals and mathematical rules called algorithms. Based on rules and algorithms, computers decide or learn how to understand data.
This research specifically used a class of machine learning algorithms called neural networks. As the name suggests, the algorithm is structured in a logical flow similar to how data is processed by neurons in the brain.
Machine learning has been used to analyze metabolomics data. However, previous studies generally did not use large-scale metabolomic data, instead scientists selected specific markers of interest and did not include data for other markers.
These limits were used because large metabolomics data typically cover thousands of different molecules, and there is a lot of variation (so-called noise) in the data. Traditional machine learning algorithms generally perform poorly with such noisy data. This is because computers have difficulty detecting meaningful patterns in random fluctuations.
The researchers’ new algorithm, CRANK-MS (which stands for Classification and Ranking Analysis Using Neural Networks, which generates knowledge from mass spectrometry), has an improved ability to classify noise and achieve complete metabolomics. We were able to use the data to provide highly accurate results.
Here we feed all the information into CRANK-MS without any data reduction in the first place. And from there you can get the model predictions and identify which metabolites are driving the predictions the most, all in one step.
“Usually, researchers using machine learning to explore correlations between metabolites and diseases first reduce the number of chemical features before feeding them into the algorithm,” said a UNSW study co-author. Dr. W. Alexander Donald said in Sydney.
“But here,” said Donald. “Without any data reduction from the start, he feeds all the information into CRANK-MS. And from there, he gets the model predictions and sees which metabolites are driving the predictions the most, all in one step.” can be identified by
Including all molecules available in the dataset “means whether there are metabolites” [molecules] We can now find things that we might have missed with traditional approaches,” said Donald.
The researchers stressed that further validation is needed to test the algorithm. However, in preliminary tests, CRANK-MS was able to distinguish between Parkinson’s and non-Parkinson’s patients with an accuracy of up to about 96%.
Notable Findings Highlight Diet and Chemical Exposure
In further analysis, the researchers identified which molecules were detected by the algorithm as being most important in identifying Parkinson’s disease.
There were some notable discoveries. For example, patients who developed Parkinson’s disease tended to have lower levels of triterpenoid chemicals known to have neuroprotective properties. This substance is found in high concentrations in foods such as apples, olives and tomatoes.
In addition, these patients often had high levels of polyfluorinated alkyl substances (PFAS), which may be markers of exposure to industrial chemicals.
“These data indicate that these metabolites are potential early indicators of PD.” [Parkinson’s disease] predates clinical PD diagnosis, is consistent with specific diets for PD prevention (such as the Mediterranean diet), and [PFASs] It may contribute to the development of PD,” the researchers wrote. The research team pointed to the need for further research on these potential biomarkers.
The scientists published the CRANK-MS algorithm for use by other researchers. The researchers say the algorithm has potential applications far beyond Parkinson’s disease.
“We built the model in a fit-for-purpose manner,” says Zhang. “What is interesting is that CRANK-MS can be easily applied to other diseases to identify new biomarkers of interest. You get results.”
