A Stacking Framework for Polygenic Risk Prediction in Admixed Individuals.

Area of Science:

• Genetics
• Bioinformatics
• Population Genetics

Background:

• Polygenic risk scores (PRS) aggregate genetic variants to estimate an individual's risk for diseases or traits.
• Current PRS methods often show reduced predictive performance in populations that differ ancestrally from the reference population used for Genome-Wide Association Studies (GWAS).
• Admixed individuals, with ancestry from multiple populations, present a unique challenge due to their continuous ancestry, making traditional population-specific PRS less effective.

Approach:

• We introduce stacking local ancestry PRS (slaPRS), a novel ensemble learning method that combines PRS derived from multiple ancestry-specific GWAS.
• slaPRS leverages local ancestry information across the genome to build more accurate risk predictions for admixed individuals.
• The method was evaluated using simulations and real-world data from African Americans and African British individuals.

Key Points:

• In simulations, slaPRS demonstrated superior performance compared to existing methods, significantly reducing ancestry-related prediction biases in African Americans.
• Analysis of lipid traits in African British individuals (UK Biobank) showed that slaPRS improved upon single-population PRS.
• slaPRS achieved performance comparable to globally combined PRS while offering a more flexible and data-driven approach.

Conclusions:

• slaPRS provides a flexible framework for integrating multiple population-specific GWAS and local ancestry information.
• This approach enhances the accuracy and reduces the ancestry dependence of PRS in admixed populations.
• slaPRS represents a significant advancement in personalized genetic risk prediction for diverse ancestral groups.