Predicting Parkinson's Disease Risk Through Protein and Peptide Level Analysis: An Evidence from EDA and Machine Learning Based Approach

Parkinson's disease (PD) is a disabling brain disorder that affects movement, cognition, and mood. Although there is no known cause or cure, it is possible to manage symptoms with treatment. PD affects about 7 million people worldwide and is projected to reach 10 million by 2030. Recent studies have explored protein and peptide levels as potential PD biomarkers that could impact Unified Parkinson's Disease Rating Scale (UPDRS) scores. In this research paper, we analyze data from 1,019 patients to predict the UPDRS score based on biomarkers. We leverage Exploratory Data Analysis (EDA) and machine learning to identify changes in protein and peptide levels that could serve as useful PD risk biomarkers. These analytical techniques utilize protein and peptide levels from general health screenings of patients for early PD detection to uncover patterns and correlations that enable accurate predictive models for UPDRS. This helps in the early detection of PD, allowing effective management and intervention. Our results show that several proteins and peptides are associated with PD risk. These proteins and peptides have a deep impact on the UPDRS score. We also find that the levels of these proteins and peptides are higher in people with PD than in healthy individuals. Finally, we obtain Symmetric Mean Absolute Percentage Error (SMAPE) scores from various machine learning algorithms to measure the accuracy of predictions. Analysis with SMAPE reveals the superiority of Random Forest as the best-performing machine learning technique for PD prediction, with a score of 0.37. In conclusion, PD is a complex disorder without a known cure. However, biomarker analysis with machine learning offers promise for early PD detection and gives clinicians and patients hope for early diagnosis and better treatment strategies.