Menu
This study is to use Machine Learning (ML) model to forecast the progression of Parkinson’s disease, through protein and peptide data measurements from Parkinson’s Disease patients. The study predicts scores of the Movement Disorder Society-Sponsored Revision of the Unified Parkinson’s Disease Rating Scale (MDS-UPDRS), which evaluates both motor and non-motor symptoms. To achieve this, a ML model is trained by learning from protein and peptide levels over time in Parkinson’s patients compared to healthy controls of similar age. This ML approach is expected to uncover critical insights into molecular changes accompanying the disease’s progression.
The ML model leverages follow-up data, alongside protein and peptide markers, to estimate the severity of Parkinson’s symptoms. These predictions extend to future periods of 6, 12, and 24 months, addressing a multi-target regression challenge that involves forecasting various outcomes simultaneously. The model’s performance is gauged using the Symmetric Mean Absolute Percentage Error (SMAPE) metric.
In this study, 248 samples were analyzed, encompassing 227 proteins and 968 peptides. Repeated analyses highlighted that while protein data generally showed lower significance in feature importance, the most critical determinant was the follow-up duration. The study contrasted experimental and control groups—the latter comprising healthy individuals and monitored biannually.
Various model-building techniques are compared, including linear regression and LightGBM, and employed SMAPE to fine-tune the training process and determine optimal iteration counts. Fusion weights for model integration were calibrated using a linear fusion method, guided by validation scores. This comprehensive approach underscores our commitment to refining prediction accuracy for Parkinson’s disease progression, leveraging ML to bridge critical gaps in our understanding of its molecular underpinnings.
