
Researchers from Boston University and UNSW Sydney have developed a machine learning tool called CRANK-MS, which can predict the onset of Parkinson’s disease before symptoms appear. The tool achieved up to 96% accuracy in forecasting disease onset in patients up to 15 years in advance by analyzing metabolomic data using neural networks. Early signs of Parkinson’s disease have been identified as unique metabolite combinations. The study emphasizes the possibility of early detection and the identification of key biomarkers. More validation and exploration are needed.
Researchers from Boston University and the University of New South Wales (UNSW Sydney) in Australia have developed a machine learning (ML) tool capable of predicting the onset of Parkinson’s disease years before the appearance of symptoms. The tool, known as CRANK-MS (Classification and Ranking Analysis using Neural Networks to Generate Knowledge from Mass Spectrometry), utilizes neural networks to analyze metabolomic data, which consists of metabolites found in human tissues and bodily fluids like blood. These metabolites can serve as biomarkers for certain diseases and conditions.
Currently, there are no blood or laboratory tests available to diagnose non-genetic Parkinson’s disease. However, by leveraging mass spectrometry (MS) to analyze metabolite profiles, researchers have discovered differences in metabolite levels in individuals who later developed Parkinson’s, even up to 15 years before clinical diagnosis. This implies that the disease could be detected much earlier than current clinical practice allows.
The research team utilized this knowledge to build their prediction model, which takes a unique approach by analyzing the entire metabolomics data. Unlike conventional statistical approaches that focus on correlations between molecules, the ML capabilities of CRANK-MS enable the researchers to explore numerous associations among the metabolites themselves. This process requires substantial computational power but allows for a comprehensive analysis of the data.
Moreover, CRANK-MS also enables researchers to analyze unedited data lists without reducing the number of chemical features beforehand. By doing so, the model provides predictions and identifies the key metabolites driving those predictions in one step. This approach allows for the potential identification of metabolites that may have been missed using traditional methods.
The researchers tested CRANK-MS on metabolomics data from 39 patients who developed Parkinson’s up to 15 years later, comparing them with a matched control group. They discovered unique combinations of metabolites that could serve as early indicators of Parkinson’s. When these combinations were used as predictors, the ML tool achieved an impressive accuracy of up to 96% in forecasting disease onset.
Dr. W. Alexander Donald, an associate professor in the School of Chemistry at UNSW Sydney, emphasized the significance of the study’s findings. He explained that the high accuracy in predicting Parkinson’s disease before a clinical diagnosis is noteworthy. Additionally, the machine learning approach allowed the researchers to identify chemical markers that play a crucial role in accurately predicting future Parkinson’s development. Some of these markers had previously been implicated in Parkinson’s disease through cell-based assays but not in human studies.
The analysis also highlighted the presence of polyfluorinated alkyl substances (PFAS) in individuals who later developed Parkinson’s. Interestingly, these same individuals exhibited lower concentrations of triterpenoids, which are neuroprotective compounds that regulate oxidative stress.
The researchers concluded that further investigations are necessary to validate CRANK-MS using larger and more diverse patient cohorts. Additionally, they emphasized the need for an in-depth exploration of the relationships between Parkinson’s disease and chemicals like PFAS and triterpenoids. Once the model is validated on larger datasets, the research team believes that CRANK-MS could be applied to other diseases to help identify new biomarkers.