Efficient polygenic risk scores for biobank scale data by exploiting phenotypes from inferred relatives

Area of Science:

• Genetics
• Bioinformatics
• Precision Medicine

Background:

• Polygenic risk scores (PRS) predict individual phenotypes but typically use unrelated individuals.
• This approach overlooks valuable genetic information from relatives.
• Integrating family data could improve PRS accuracy and utility.

Purpose of the Study:

• To evaluate the predictive performance of PRS using a smaller cohort that includes first-degree relatives compared to large cohorts of unrelated individuals.
• To determine if family information enhances PRS accuracy for various traits, particularly lifestyle-related ones.

Main Methods:

• Utilized UK Biobank data for 50 different traits.
• Compared prediction accuracy of PRS models using 5,000 individuals with first-degree relatives against models using 220,000 unrelated individuals.
• Analyzed prediction accuracy specifically for lifestyle traits.

Main Results:

• A cohort of 5,000 individuals with relatives achieved prediction accuracy comparable to 220,000 unrelated individuals for 50 traits (mean accuracy 0.26 vs. 0.24).
• For lifestyle traits, the smaller cohort including relatives showed significantly higher prediction accuracy (0.22 vs. 0.16).
• This demonstrates a 44-fold reduction in sample size with similar or improved predictive power.

Conclusions:

• Integrating first-degree relative information into polygenic prediction models can achieve high accuracy with substantially smaller sample sizes.
• This family-inclusive approach shows particular promise for improving prediction of lifestyle traits.
• Findings support the acceleration of precision health and clinical interventions through family-based polygenic prediction.