A Machine Learning Approach to Parkinson's Disease Blood Transcriptomics.

Area of Science:

• Biomarkers
• Neurodegenerative Diseases
• Genomics

Background:

• Parkinson's disease (PD) presents a significant health burden with delayed diagnosis.
• Current diagnostic methods often identify PD after irreversible neurological damage has occurred.
• Effective early diagnostic tools are crucial for timely intervention and management.

Purpose of the Study:

• To evaluate blood transcriptomics for early Parkinson's disease (PD) diagnosis.
• To identify candidate genes associated with PD using machine learning.
• To explore the functional significance of identified genes in PD pathogenesis.

Main Methods:

• Utilized transcriptome data from approximately 550 early PD subjects and healthy controls (HC) from the Parkinson's Progression Markers Initiative (PPMI) study.
• Employed a nested feature selection procedure utilizing Random Forests and XGBoost algorithms for classification.
• Conducted functional analysis using Gene Ontology (GO) and KEGG pathways to interpret candidate genes.

Main Results:

• Achieved an Area Under the Curve (AUC) of 72% in classifying PD versus HC.
• Identified 493 candidate genes potentially associated with early Parkinson's disease.
• Functional analysis provided insights into the biological pathways implicated by these candidate genes.

Conclusions:

• Blood transcriptomics is a promising approach for the early detection of Parkinson's disease.
• The identified 493 candidate genes offer potential biomarkers for PD diagnosis.
• Further research into these genes could elucidate PD mechanisms and inform therapeutic strategies.