Machine Learning for Parkinson's Disease Diagnosis Using proteomics Data from Human Tears

Parkinson's disease is a neurodegenerative condition that has a significant impact on the quality of life of affected individuals. Despite advances in medicine, the difficulty of diagnosis remains a challenge, with only about 4% of cases being diagnosed before the age of 50. Therefore, the discovery of new non-invasive diagnostic methods is important for early detection and monitoring of the disease. In this study, molecular data from human tear samples obtained from proteomic studies were used in conjunction with machine learning to develop a promising approach for diagnosing Parkinson's disease. Random Forest and Support Vector Machine-based classification models were compared, identifying important variables for classification that may aid in the search for biomarkers. Moreover, with an accuracy of 93.75%, the Random Forest model proved highly effective in-patient classification, while the SVM model had an accuracy of 75%. In conclusion, this study may be an important step in the development of new non-invasive and accurate diagnostic strategies for Parkinson's disease.

The data used for building the classifiers were obteined in ProteomeXchange identifier (10.6019/PXD028811).