Predictive Modeling of Parkinson’s Disease Progression Through Proteomic and Peptidomic Analysis

Parkinson’s disease (PD) is a disabling brain disorder that affects normal functions. To assess the disease progression, the Unified PD Rating Scale (UPDRS) is commonly used. UPDRS includes four components: intellectual function, mood and behavior; activities of daily living; motor examination and motor complications. The present work analyzes a dataset of PD patients that includes UPDRS scores of each clinical stage at different visits, and laboratory data such as protein and peptide abundance values derived from mass spectrometry readings from cerebrospinal fluid (CSF) samples. Data was obtained from the AMP®-Parkinson’s Disease Progression Prediction public database. The objective is to predict the UPDRS scores of the different stages of PD as well as to identify relevant biomarkers of the disease progression using machine learning algorithms. For this purpose, Extreme Gradient Boosting (XGBoost) and Light Gradient Boosting (LightGBM) algorithms were used. Additionally, the resulting predictions from basic regression models (Linear Regression, K-Nearest Neighbors Regression, Support Vector Regression, and Stochastic Gradient Descent Regressor) were incorporated to ensemble both learning algorithms and achieve better predictive performance. SMAPE was used as an accuracy metric, obtaining a score of 55.55 for LightGBM and 50.63 for XGBoost. Among the main determining factors to predict UPDRS scores, proteins with documented alterations in PD patients have been found. The following proteins were also identified as important in the prediction: NCHL1, RNT2, ISLR, and B4GAT1. These findings contribute to obtain a better understanding of the determination of biomarkers to predict the progression of Parkinson’s disease.