An explainable ensemble machine learning model using baseline blood transcriptomics to predict Parkinson's disease...

Area of Science:

• Neuroscience
• Genetics
• Computational Biology

Background:

• Predicting Parkinson's disease (PD) motor progression is difficult with current neuroimaging techniques.
• Blood-based transcriptomic profiling offers a more accessible and cost-effective alternative for PD research.
• Identifying reliable biomarkers for PD progression is crucial for effective patient management.

Purpose of the Study:

• To develop and validate a machine learning model using blood transcriptomic data to predict 12-month motor severity in PD.
• To identify key transcriptomic features and biological pathways associated with PD motor progression.
• To explore the prognostic value of specific PD risk genes and cellular pathways.

Main Methods:

• A Stacking Regressor ensemble model was built using baseline data from the Parkinson's Progression Markers Initiative (PPMI) cohort (n=390).
• The model integrated blood RNA sequencing (RNA-seq) and clinical data to predict 12-month UPDRS Part III scores.
• SHapley Additive exPlanations (SHAP) analysis was used to identify prognostic features and pathway contributions.

Main Results:

• The model achieved an R² of 0.551 and MAE of 6.01 on an independent test set (n=78).
• The UPDRS baseline by PINK1 gene interaction was the most influential predictive feature.
• VPS35, GBA, and LRRK2 genes were prominent transcriptomic features, with mitochondrial dysfunction showing the highest pathway contribution.

Conclusions:

• Machine learning integrating blood transcriptomics and clinical data effectively predicts PD motor progression.
• The interplay between initial clinical status and genetic factors, especially PINK1, significantly impacts prognosis.
• Mitochondrial dysfunction is a dominant prognostic signal in PD, suggesting key targets for future research and therapeutics.