AI-Based Amino Acid Profiling: Predictive Modeling of LNAA Metabolism in Parkinson’s Disease

Abstract

Parkinson’s disease (PD) is a progressive neurodegenerative disorder characterized by dopaminergic neuronal loss and systemic metabolic alterations. Large neutral amino acids (LNAAs), including tyrosine, threonine, phenylalanine, leucine, isoleucine, and valine, are closely involved in dopamine biosynthesis and cerebral amino acid transport. This study aimed to evaluate the classification performance of plasma LNAA profiles in PD using integrated statistical analyses and artificial intelligence (AI)–based machine learning (ML) approaches. Plasma samples from 47 patients with PD and 43 age- and sex-matched healthy controls were analyzed using liquid chromatography–tandem mass spectrometry (LC–MS/MS) to quantify LNAA concentrations. Data normality was assessed using the Shapiro–Wilk test, and between-group comparisons were performed using the Mann–Whitney U test. Univariate classification performance was evaluated by receiver operating characteristic (ROC) analysis. Multivariate ML models, including Random Forest (RF), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost), were trained to distinguish PD patients from controls. Model performance was assessed using accuracy, ROC–AUC, precision, and recall metrics. Logistic regression analysis was applied to identify multivariate linear associations among selected amino acids. Univariate analyses showed that threonine and tyrosine levels differed significantly between PD patients and controls (p < 0.05), whereas leucine and isoleucine exhibited substantial overlap and limited individual classification ability. In contrast, multivariate logistic regression identified threonine, leucine, and isoleucine as significant predictors of PD status (p < 0.01). Ensemble ML models further highlighted the importance of these metabolites. RF and XGBoost achieved the highest classification performance (ROC–AUC = 0.96), while SVM demonstrated slightly lower but robust performance (ROC–AUC = 0.88). AI-assisted multivariate modeling of plasma LNAA profiles enables effective classification of Parkinson’s disease by capturing coordinated metabolic patterns that are not evident in univariate analyses. This integrated biochemical–AI approach offers a promising framework for metabolic phenotyping and risk stratification in PD.