# Machine learning-guided analysis of metabolomic alterations in Parkinson’s disease with comorbid symptoms

## TL;DR

This study uses machine learning and metabolomics to identify biomarkers for Parkinson’s disease and its comorbidities, offering insights into metabolic changes.

## Contribution

The novel use of machine learning to identify metabolomic biomarkers for Parkinson’s disease and its non-motor comorbidities.

## Key findings

- 2,601 metabolites were detected in blood plasma samples from Parkinson’s patients and controls.
- Machine learning models effectively distinguished PD from healthy individuals and identified comorbidities like RBD and insomnia.
- Metabolic biomarkers may help understand disease progression and symptom associations in PD.

## Abstract

As a common neurodegenerative disorder, Parkinson’s disease (PD) primarily affects dopaminergic neurons, leading to progressive motor disabilities along with a spectrum of non-motor complications. The early identification of Parkinson’s disease, as well as the exploration of biomarkers related to its associated comorbidities, remains an important focus of current research.

In this study, a metabolomics approach combined with machine learning techniques was applied to explore potential biomarkers for PD and its related comorbid conditions. Using liquid chromatography–tandem mass spectrometry (LC–MS/MS), blood plasma samples were analyzed from individuals with PD, PD with rapid eye movement sleep behavior disorder (PD+RBD), PD with insomnia (PD + insomnia), and healthy controls, resulting in the detection of 2,601 metabolites. Multivariate statistical methods—including the unsupervised principal component analysis (PCA) and the supervised techniques of partial least squares discriminant analysis (PLS-DA) and orthogonal partial least squares discriminant analysis (OPLS-DA)—were employed to investigate intergroup metabolic variations. Machine learning algorithms, such as recursive feature elimination in conjunction with logistic regression, random forest, and support vector machines, were used to assist in selecting discriminative metabolites and constructing classification models.

These models showed strong internal performance in distinguishing PD from healthy individuals and in characterizing PD patients with non-motor comorbidities such as RBD and insomnia. Overall, the results suggest that metabolic biomarkers may provide valuable insights into disease-related and symptom-associated metabolic alterations in Parkinson’s disease.

This study provides a basis for future investigations aimed at validating these findings and further exploring their potential relevance in clinical research.