Kinship estimation bias carries over to heritability estimation bias using variance components

Area of Science:

• Population Genetics
• Quantitative Genetics
• Statistical Genomics

Background:

• Heritability estimation is crucial for understanding genetic contributions to traits and diseases.
• Variance components models rely on kinship matrices for heritability estimation.
• Standard kinship estimators like GCTA's can be biased in structured populations.

Purpose of the Study:

• To characterize heritability estimation biases in GCTA arising from kinship estimation errors in structured populations.
• To evaluate the performance of different kinship estimators under population structure.

Main Methods:

• Derivation of a closed-form expression for heritability bias in the ratio-of-means (ROM) estimator.
• Simulation studies using admixture and family structures with 1000 Genomes genotypes.
• Analysis of real-world phenotype data from three structured cohorts.

Main Results:

• The standard mean-of-ratios (MOR) estimator exhibits biases due to shared factors with ROM and additional bias from low-frequency variants.
• The Popkin estimator demonstrated unbiased heritability estimates across simulated and real structured datasets.
• Pedigree-only estimates showed upward heritability biases in structured populations.
• The MOR estimator produced variable biases (downward or upward) depending on population structure and variant frequency.

Conclusions:

• Common kinship estimators introduce significant biases in heritability estimates for structured populations.
• The Popkin estimator is a reliable alternative for unbiased heritability estimation in the presence of population structure.
• Accurate kinship estimation is essential for valid heritability inference in diverse populations.