Pitfalls in performing genome-wide association studies on ratio traits

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Genome-wide association studies (GWAS) frequently analyze trait ratios (e.g., body mass index) to implicitly adjust for numerator-denominator associations.
• However, analyzing ratios does not statistically condition on the denominator, leading to ambiguity in interpreting genetic associations.

Purpose of the Study:

• To investigate the validity of performing GWAS on trait ratios.
• To determine if associations found in ratio GWAS are attributable to the numerator, denominator, or both.
• To compare ratio GWAS with an adjusted model that conditions on the denominator.

Main Methods:

• Analysis of 10 common ratio traits using GWAS.
• Comparison of results from ratio GWAS and an adjusted GWAS model (numerator GWAS conditioning on denominator).
• Examination of body mass index (BMI) and waist-to-hip ratio (WHR) as case studies.
• Application of a novel method for correcting genetic collider bias using leave-one-chromosome-out polygenic scoring.

Main Results:

• Associations in ratio GWAS can be entirely driven by the denominator trait.
• The ratio model and the adjusted model showed disagreement at approximately one-third of loci across 10 common ratio traits.
• Variants unique to the ratio model were more strongly associated with the denominator (e.g., height for BMI), while variants unique to the adjusted model were more strongly associated with the numerator (e.g., weight for BMI).
• The adjusted model, while offering clearer interpretations, is susceptible to collider bias.

Conclusions:

• Performing GWAS on trait ratios can lead to spurious associations driven by the denominator.
• An adjusted model (GWAS on numerator, conditioning on denominator) provides clearer interpretations but requires correction for collider bias.
• A validated method using polygenic scoring is proposed to correct for the genetic component of collider bias in adjusted models.