Abstract
Polygenic risk scores are emerging as a potentially powerful tool to predict future phenotypes of target individuals, typically using unrelated individuals, thereby devaluing information from relatives. Here, for 50 traits from the UK Biobank data, we show that a design of 5,000 individuals with first-degree relatives of target individuals can achieve a prediction accuracy similar to that of around 220,000 unrelated individuals (mean prediction accuracy = 0.26 vs. 0.24, mean fold-change = 1.06 (95% CI: 0.99-1.13), P-value = 0.08), despite a 44-fold difference in sample size. For lifestyle traits, the prediction accuracy with 5,000 individuals including first-degree relatives of target individuals is significantly higher than that with 220,000 unrelated individuals (mean prediction accuracy = 0.22 vs. 0.16, mean fold-change = 1.40 (1.17-1.62), P-value = 0.025). Our findings suggest that polygenic prediction integrating family information may help to accelerate precision health and clinical intervention.
Introduction
In the field of Genome-Wide Association Studies (GWAS), relatives have traditionally been viewed as "nuisance variables." To cleanly estimate heritability—the proportion of trait variation attributable to genetics—researchers typically exclude close relatives to avoid confounding genetic signals with shared family environments.
However, for clinical prediction via Polygenic Risk Scores (PRS), the objective shifts from theoretical estimation to predictive accuracy. By ignoring relatives, current models discard valuable information regarding shared genetic segments and shared environments. This study argues that family data is a massive, underutilized resource that can significantly enhance the potency of risk scores.
Methodology and Study Design
The researchers compared two distinct study designs using data from the UK Biobank to evaluate the efficiency of family-based prediction:
- Large-scale design: Utilizing approximately 220,000 unrelated individuals.
- Small-scale design: Utilizing only 5,000 individuals, specifically selecting those with first-, second-, or third-degree relatives of the target individuals.
Theoretical Framework: $M_e$ and HBLUP
The efficiency of including relatives is rooted in the concept of $M_e$ (the effective number of chromosome segments). Because relatives share large segments of DNA, the statistical "search space" for the model is reduced, allowing for more accurate effect estimation with smaller samples.
To maximize this information, the study employed HBLUP (H-matrix Best Linear Unbiased Prediction). This technique, borrowed from livestock breeding, allows for the integration of phenotypes from relatives who may not have been genotyped, using only their known pedigree relationship to the target individual. This approach captures both genetic proximity and shared "cultural inheritance."
Key Insights and Results
The Power of Proximity
The study revealed a remarkable 44-fold increase in efficiency. Including first-degree relatives in a sample of just 5,000 individuals provided a mean prediction accuracy of 0.26, slightly outperforming the massive sample of 220,000 unrelated individuals (0.24) across 50 complex traits.
Lifestyle Synergy
The boost was most dramatic for lifestyle traits, such as smoking status and education level. For these traits, the small-scale design with relatives was significantly superior to the large-scale unrelated design (accuracy of 0.22 vs. 0.16). This suggests that family members share habits and socioeconomic factors that reinforce the genetic signal, a phenomenon the authors term "lifestyle synergy."
Clinical Impact
Beyond raw accuracy, the family-informed design showed greater clinical utility. The small-scale design achieved a higher Odds Ratio (1.81) for identifying individuals in the top 10% of risk compared to the large-scale unrelated design (1.62), demonstrating its potential for more effective patient stratification.
Limitations and Future Directions
While the results are compelling, the study notes several constraints. The UK Biobank currently has a limited number of close relatives per participant; as biobanks grow to include more complete nuclear families, accuracy is expected to rise further. Additionally, the study focused on individuals of European ancestry, necessitating further research into diverse global populations. Finally, the HBLUP method is computationally demanding, requiring more optimized algorithms to scale to millions of participants. Despite these hurdles, the integration of family information represents a significant step forward for precision health.
Summary updated