Predictive Big Data Analytics: A Study of Parkinson's Disease Using Large, Complex, Heterogeneous, Incongruent,...

Area of Science:

• Neuroscience
• Data Science
• Biomedical Informatics

Background:

• The Parkinson's Progression Markers Initiative (PPMI) provides a unique Big Data archive for Parkinson's Disease (PD).
• Integrating complex, heterogeneous Big Data presents challenges for traditional data analysis methods.
• Previous studies explored PD risk factors including genetics, environment, and lifestyle.

Purpose of the Study:

• To develop and validate model-based and model-free approaches for PD classification and prediction using Big Data.
• To jointly process complex PPMI imaging, genetics, clinical, and demographic data for PD risk exploration.

Main Methods:

• Developed a comprehensive protocol for end-to-end data characterization, manipulation, processing, cleaning, analysis, and validation.
• Introduced methods for rebalancing imbalanced cohorts.
• Utilized a wide spectrum of classification methods, focusing on machine-learning based approaches.

Main Results:

• Model-free machine-learning classification methods achieved over 96% accuracy, sensitivity, and specificity in predicting PD in PPMI subjects.
• Clinical (UPDRS scores), demographic (age), genetic, and neuroimaging biomarker data contributed to predictive analytics.
• Model-based predictive approaches failed to generate accurate and reliable diagnostic predictions.

Conclusions:

• Model-free Big Data machine learning methods outperform model-based techniques for PD diagnosis forecasting.
• Statistical rebalancing of cohort sizes improves discrimination for predictive analytics.
• The developed methods, software, and protocols are openly shared for studying other neurodegenerative disorders.