An efficient LASSO framework for admixture-aware polygenic scores.

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Polygenic scores (PGSs) are valuable for clinical applications but often perform poorly in non-European populations due to training data bias.
• Existing cross-population PGS methods struggle with individuals of recent admixture, limiting their generalizability.
• The GAUDI method addresses admixture but is computationally intensive and limited to two-way admixture.

Purpose of the Study:

• To introduce HAUDI, an efficient LASSO-based framework for constructing polygenic scores in admixed populations.
• To overcome the computational limitations and multi-way admixture constraints of the GAUDI method.
• To enhance the accuracy and portability of PGSs across diverse ancestral backgrounds.

Main Methods:

• HAUDI reparametrizes the GAUDI model into a standard LASSO problem for improved computational efficiency.
• The framework is extended to accommodate multi-way admixture settings.
• Performance was evaluated through extensive simulations and real-world data applications across 18 clinical phenotypes.

Main Results:

• HAUDI demonstrates comparable or superior performance to GAUDI in simulations, with significantly reduced computation time.
• In real data, HAUDI uniformly outperforms GAUDI across 18 clinical phenotypes, including triglycerides (TG), C-reactive protein (CRP), and mean corpuscular hemoglobin concentration (MCHC).
• HAUDI shows substantial benefits over ancestry-agnostic PGSs for traits like white blood cell count (WBC) and chronic kidney disease (CKD), and outperforms SDPR_admix.

Conclusions:

• HAUDI provides an efficient and accurate method for constructing polygenic scores in admixed populations.
• The framework significantly improves PGS performance and generalizability across diverse ancestries.
• HAUDI represents a substantial advancement for personalized medicine and genetic risk prediction in global populations.