Partitioning Fraction of Variance Explained into Strong Localized Effects and Weak Diffuse Effects

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• High-dimensional genetic data pose challenges for estimating the fraction of variance explained (FVE) by single-nucleotide polymorphisms (SNPs).
• Standard heritability estimators assume Gaussian SNP effect sizes, but empirical data show heavy-tailed distributions, violating the bounded-kurtosis effect (BKE) condition.
• This violation leads to biased FVE estimates in widely used methods like GWASH and LDSC regression.

Purpose of the Study:

• To develop a robust framework for estimating FVE that accommodates heavy-tailed and heterogeneous SNP effect distributions.
• To partition total heritability into contributions from strong and weak genetic effects.
• To introduce a test for BKE condition violations and compare screening procedures for strong-effect SNPs.

Main Methods:

• A decomposed FVE estimation framework partitioning heritability into strong and weak genetic effects.
• Utilizing low-dimensional adjusted R-squared for strong effects and an extended FVE methodology for weak effects under BKE compliance.
• Developing a BKE violation detection test and evaluating high-dimensional screening procedures.

Main Results:

• The proposed decomposition significantly enhances estimation accuracy compared to existing methods when heavy-tailed effects are present.
• Simulation studies confirm improved accuracy in the presence of effect heterogeneity.
• Application to the ABCD Study provided more reliable heritability estimates for the PolyVoxel Score.

Conclusions:

• Accommodating SNP effect heterogeneity is crucial for accurate heritability estimation in large-scale genomic studies.
• The developed method offers a more reliable approach to estimating the fraction of variance explained by SNPs.
• This framework improves the understanding of genetic architecture, particularly for complex traits and biomarkers.