Interaction-Based Feature Selection Algorithm Outperforms Polygenic Risk Score in Predicting Parkinson's Disease...

Area of Science:

• Genetics
• Machine Learning
• Computational Biology

Background:

• Polygenic risk scores (PRS) are limited in predicting complex diseases due to their inability to account for gene-gene interactions.
• Machine learning (ML) offers potential for identifying these interactions to improve predictive models.

Purpose of the Study:

• To develop and evaluate an ML-based approach for predicting Parkinson's disease (PD) susceptibility by incorporating gene-gene interactions.
• To demonstrate the superiority of interaction-based models over traditional PRS.

Main Methods:

• A data-mining preprocessing step was used to reduce features and enable ML algorithms to identify gene-gene interactions.
• The approach was applied to the Parkinson's Progression Markers Initiative (PPMI) dataset.
• An interaction-based prediction model was compared against PRS using Area Under the Curve (AUC).

Main Results:

• The interaction-based prediction model achieved an AUC of 0.85, significantly outperforming PRS (AUC = 0.58).
• Feature importance analysis provided insights into PD mechanisms, highlighting interactions between genes like TMEM175 and GAPDHP25.

Conclusions:

• Interaction-based ML models offer improved genetic prediction for complex diseases like PD.
• This methodology may help resolve the 'missing heritability' issue in genetic studies.