Abstract: In the genome-wide association analysis of population-based biobanks, most diseases have low prevalence, which results in low detection power. One approach to tackle the problem is using family disease history, yet existing methods are unable to address type I error inflation induced by increased correlation of phenotypes among closely related samples, as well as unbalanced phenotypic distribution. We propose a new method for genetic association test with family disease history, TAPE (mixed-model-based Test with Adjusted Phenotype and Empirical saddlepoint approximation), which controls for increased phenotype correlation by adopting a two-variance-component mixed model, accounts for case-control imbalance by using empirical saddlepoint approximation, and is flexible to incorporate any existing adjusted phenotypes such as phenotypes from the LT-FH method. We show through simulation studies and analysis of UK-Biobank data of white British samples and the Korean Genome and Epidemiology Study (KoGES) of Korean samples that the proposed method is robust and yields better calibration compared to existing methods while gaining power for detection of variant-phenotype associations.
Introduction
Genome-wide association studies (GWAS) using large-scale population biobanks have revolutionized our understanding of the genetic architecture of complex diseases. However, many conditions—such as Parkinson’s disease or specific cancers—exhibit low prevalence within these cohorts. This rarity leads to a significant loss of statistical power, making it difficult to identify associated genetic variants using standard case-control designs.
One promising strategy to overcome this is the use of family disease history. By treating healthy individuals with a strong family history of a disease as "proxy cases," researchers can effectively increase the number of cases in their analysis. However, existing methods like LT-FH (Liability Threshold Model with Family History) face two critical challenges:
- Relatedness Inflation: In biobanks with many related individuals (e.g., siblings), shared family history creates high correlation in proxy phenotypes. Standard models fail to account for this specific correlation, leading to inflated Type I error rates (false positives).
- Case-Control Imbalance: When the ratio of cases to controls is highly skewed, the mathematical assumption that data follows a normal distribution breaks down at the "tails," resulting in poorly calibrated p-values for rare genetic variants.
The TAPE Framework
To address these issues, we developed TAPE (mixed-model-based Test with Adjusted Phenotype and Empirical saddlepoint approximation). TAPE is a robust statistical framework designed to leverage family history while maintaining strict control over false positives, even in the presence of related samples.
Methodological Innovations
The TAPE framework operates through a multi-step process:
- Phenotype Adjustment: TAPE calculates a latent disease risk for each participant. It is flexible, supporting simple weighted proportions (TAPE-WP) or integrating with existing models like LT-FH (TAPE-LTFH).
- Two-Variance-Component Mixed Model: This is the core of TAPE's ability to handle relatedness. While standard models use a Genetic Relatedness Matrix (GRM) to capture general background similarity, TAPE adds a second, sparse kinship matrix. This acts as a surgical tool to account for the heightened correlation between close relatives who share identical parental disease histories.
- Empirical Saddlepoint Approximation (ESPA): To manage extreme case-control imbalance, TAPE utilizes ESPA. Instead of assuming a standard distribution, ESPA uses the actual residuals from the data to estimate the distribution's tails, ensuring accurate p-values even for rare variants in skewed datasets.
To ensure the method is scalable to biobank-sized datasets with millions of variants, the implementation utilizes high-performance computational techniques, including the Preconditioned Conjugate Gradient (PCG) algorithm.
Experimental Validation
The performance of TAPE was evaluated using simulated data and real-world datasets from the UK Biobank (UKB) and the Korean Genome and Epidemiology Study (KoGES).
UK Biobank Analysis
We analyzed 10 binary traits in approximately 409,000 white British individuals. For diseases with low prevalence, such as Parkinson’s disease (case-control ratio ~1:360), previous methods like LT-FH showed significant p-value inflation in QQ-plots—a clear sign of false positives. In contrast, TAPE’s results remained perfectly calibrated.
Furthermore, TAPE demonstrated a dramatic increase in discovery power. For example, TAPE-LTFH identified 726 genome-wide significant variants across the studied traits, more than doubling the 344 variants found by SAIGE, a standard GWAS method that does not incorporate family history.
KoGES Analysis
In the KoGES dataset (~72,000 individuals), TAPE was applied to gastric cancer and diabetes. For gastric cancer, TAPE identified three significant variants that were missed by standard GWAS. These variants had been previously linked to gastric cancer in other Asian cohorts, providing strong real-world validation of TAPE’s ability to uncover "hidden" genetic signals.
Discussion and Future Directions
TAPE represents a significant advancement in GWAS methodology, allowing researchers to fully utilize biobank data without excluding related individuals to avoid bias. By correcting for both relatedness inflation and case-control imbalance, it provides a more powerful and reliable tool for genetic discovery.
However, some limitations remain. The reliance on self-reported family history may introduce bias compared to clinical records. Additionally, the model assumes an infinitesimal genetic architecture; while robust, it may not be optimal for traits dominated by a few large-effect variants. Future work will explore extending TAPE to ordinal phenotypes (e.g., disease severity levels) to further enhance its statistical power.
Summary updated