Metabolomic signatures in late-onset Parkinson's disease: a cross-sectional analysis of serum biomarkers in the Tracking Parkinson's cohort

Thesis

Parkinson’s disease (PD) is the second most common neurodegenerative disease, affecting 1% of the population over 60 years of age (Xiao et al. 2025). Currently, there is no single accepted diagnostic test for PD, and the need for accurate, earlier diagnostic biomarkers remains critical (Ben-Shlomo et al. 2024). Metabolomics offers a promising approach for biomarker studies, as the metabolome represents the final downstream product of biological processes and provides a detailed snapshot of an organism’s biochemical state (Clish 2015). This study analyzed baseline serum metabolomic data from the Tracking PD cohort (Malek et al. 2015), comprising 1,482 late-onset PD patients (disease duration <3.5 years, diagnosed ≥50 years) and 203 genetically-related unaffected sibling controls, with the aims of characterising metabolic alterations in PD serum samples, developing and validating metabolite- based classification models, and identifying specific metabolites and biological pathways most significantly implicated in PD pathology. Multiple analytical approaches were employed, including unsupervised clustering (Principal Component Analysis [PCA] and Uniform Manifold Approximation and Projection [UMAP]), generalised linear modelling (GLM), differential abundance analysis (Linear Models for Microarray Data [LIMMA]), and machine learning classification with SHapley Additive exPlanations (SHAP) feature importance analysis. Twenty-one metabolites were consistently altered across multiple statistical methods (LIMMA, GLM, SHAP), with fatty acid amides and tyrosine metabolism-associated metabolites playing central roles in distinguishing PD from healthy control metabolic profiles. Machine learning models achieved an area under the receiver operating characteristic curve (AUC) of 0.9884 for PD classification. Primary fatty acid amides (PFAMs) were elevated in both treated and untreated PD patients, suggesting these alterations reflect underlying disease mechanisms rather than treatment effects. These findings identified widespread pathway-level disruptions in PD serum samples, with PFAMs representing a potential therapeutic target. Limitations of this project lie in the large class imbalance inherent to the samples, and future directions would be to further investigate PFAMs in other PD and related neurodegenerative cohorts.

Authors: Loison, N

• DOI: 10.5287/ora-azqa6aajr
• Type of award: MSc by Research
• Level of award: Masters
• Awarding institution: University of Oxford
• Language: English
• Keywords: late-onset; machine learning; PD; ML; serum; metabolomics; bioinformatics; Parkinson's Disease; Primary fatty acid amides; biomarkers; untargeted mass spectrometry
• Subjects: Medical sciences; Neurosciences; Bioinformatics; Psychiatry