PNL: a software to build polygenic risk scores using a super learner approach based on PairNet, a Convolutional...

Area of Science:

• Genetics and Bioinformatics
• Computational Biology
• Precision Medicine

Background:

• Polygenic risk scores (PRSs) are valuable for disease prediction but often have limited discriminative power in general populations.
• Existing PRS modeling techniques require testing to determine optimal methods for specific applications.
• Developing robust PRS models is crucial for advancing early diagnosis and personalized treatment strategies.

Purpose of the Study:

• To introduce PNL, a novel tool for building optimized PRS models by integrating candidate models.
• To enhance PRS predictive performance using target population data and/or trans-ethnic approaches.
• To evaluate PNL's effectiveness in improving disease prediction accuracy compared to individual methods.

Main Methods:

• PNL utilizes the PairNet algorithm, a computationally efficient Convolutional Neural Network.
• The tool integrates multiple PRS candidate models, leveraging population-specific and trans-ethnic data.
• Case studies were conducted for asthma, type 2 diabetes, and vertigo using Taiwan Biobank (TWB) and UK Biobank (UKBB) data.

Main Results:

• PNL models using only TWB data matched or surpassed individual methods' Area Under the Curves (AUCs) for asthma, type 2 diabetes, and vertigo.
• Incorporating UKBB data further improved PNL's performance for asthma and type 2 diabetes.
• For vertigo, UKBB data did not enhance PNL's AUC, indicating that model integration does not always guarantee improved performance and helps mitigate overfitting concerns.

Conclusions:

• PNL offers a versatile approach to optimize PRS models, enhancing their predictive power for complex diseases.
• The tool demonstrates the potential to improve early disease diagnosis and guide personalized treatment strategies.
• PNL's flexible integration of diverse data sources and modeling techniques represents a significant advancement in PRS development.