Representation in genetic studies affects inference about genetic architecture.

Area of Science:

• Population Genetics
• Genomic Epidemiology
• Statistical Genetics

Background:

• Understanding genetic architecture is key to evolutionary and biological insights.
• Genetic architecture inferences rely on heterogeneous cohort data.
• Biobank recruitment strategies (e.g., UK Biobank vs. FinnGen) create data distribution differences.

Purpose of the Study:

• To investigate how biobank representation influences genetic architecture inferences.
• To assess the impact of cohort heterogeneity on genetic architecture summaries.
• To identify factors driving variations in genetic architecture estimates across biobanks.

Main Methods:

• Utilized Genome-Wide Association Study (GWAS) data from UK Biobank (UKB), FinnGen, and All of Us (AoU).
• Compared summaries of genetic architecture, including effective polygenicity, SNP heritability, and sign bias.
• Employed statistical simulations and data analysis to test hypotheses regarding trait distribution skewness.

Main Results:

• Effective polygenicity showed minimal variation across biobanks.
• SNP heritability was consistently lower in disease-enriched (AoU) versus population-based (UKB) biobanks.
• Inferred sign bias varied considerably, explained primarily by trait distribution skewness (81-97% variance explained).

Conclusions:

• Biobank representation significantly affects inferences about genetic architecture, particularly SNP heritability and sign bias.
• Trait distribution skewness is a major driver of observed sign bias, independent of other study characteristics.
• Careful consideration of biobank characteristics and trait distributions is essential for accurate genetic architecture interpretation.