Machine learning model based on plasma proteomics for the identification of Parkinson's disease

Area of Science:

• Neuroscience
• Biochemistry
• Computational Biology

Background:

• Reliable biomarkers are crucial for differentiating Parkinson's disease (PD) from other neurological conditions.
• High-throughput proteomics and machine learning (ML) offer advanced tools for biomarker discovery.

Purpose of the Study:

• To develop and validate a plasma protein-based biomarker panel for diagnosing Parkinson's disease using ML.
• To identify predictive proteins and elucidate biological pathways associated with PD.

Main Methods:

• Plasma proteomic profiles were analyzed from 698 participants (PD cases, controls, other neurological conditions) using the Olink Explore 3072 assay.
• Differential protein abundance, pathway enrichment, and Boruta algorithm were used to identify predictive proteins.
• A stacking ensemble ML model was trained on 11 proteins and validated in independent cohorts using SHAP and network analyses.

Main Results:

• The ML model achieved high diagnostic accuracy (AUC 0.939 in Test Set) and validated across four independent cohorts.
• Eleven proteins (APOH, ARG1, CCN1, CXCL1, CXCL8, DDC, GRAP2, IL1RAP, OSM, PRL, SPRY2) were identified as key predictive features.
• Network and pathway analyses revealed associations with inflammation, ErbB signaling, T-cell receptor signaling, and lipid metabolism.

Conclusions:

• Plasma protein biomarkers combined with ML show significant potential for accurate and reliable Parkinson's disease diagnosis.
• The developed model enhances clinical utility and deepens the biological understanding of PD.
• This approach may help identify novel risk factors and pathways involved in Parkinson's disease pathogenesis.