Accurate and Efficient Estimation of Local Heritability using Summary Statistics and LD Matrix

Area of Science:

• Quantitative genetics
• Statistical genomics
• Bioinformatics

Background:

• Existing methods for estimating SNP-heritability from GWAS summary statistics are less efficient than REML using individual-level data.
• Improving heritability estimator precision is crucial for regional analyses where local genetic variances are often small.

Approach:

• Introduced HEELS, a novel estimator for local heritability utilizing summary statistics (Z-scores) and the empirical LD matrix.
• Developed a unified framework to evaluate LD approximation strategies, proposing a low-rank plus banded matrix representation.
• Reduced computational and memory costs associated with LD matrix usage.

Key Points:

• HEELS achieves statistical efficiency comparable to REML (over 92% relative efficiency) using only summary-level data.
• HEELS significantly outperforms existing summary-statistics-based methods like GRE and LDSC in heritability estimation precision.
• The proposed LD approximation enhances computational efficiency and reduces resource requirements for HEELS.

Conclusions:

• HEELS offers a statistically efficient alternative for heritability estimation using summary statistics.
• The LD approximation strategies improve the practicality and performance of genetic analyses.
• Demonstrated HEELS's effectiveness and LD approximation advantages through simulations and UK Biobank data analysis.