Gene Level Meta-Analysis of Quantitative Traits by Functional Linear Models.

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Meta-analysis of genetic data presents challenges due to heterogeneity across studies (e.g., designs, markers, covariates).
• Existing methods like MetaSKAT may not fully capture complex genetic associations.
• Developing advanced statistical methods is crucial for robust genetic meta-analysis.

Purpose of the Study:

• To develop novel functional linear models for genetic meta-analysis.
• To introduce likelihood-ratio test (LRT) and F-distributed statistics for association testing.
• To evaluate the performance of these new methods against existing ones.

Main Methods:

• Functional linear models were developed to link genetic data with quantitative traits, adjusting for covariates.
• Likelihood-ratio test (LRT) and F-distributed statistics were introduced for association testing of multiple variants within a genetic region.
• Extensive simulations were conducted to assess type I error rates and statistical power.

Main Results:

• The proposed LRT and F-distributed statistics demonstrated excellent control of type I error rates.
• These new methods exhibited higher statistical power compared to the meta-analysis sequence kernel association test (MetaSKAT).
• Application to European blood lipid data revealed more significant associations and smaller P-values than MetaSKAT.

Conclusions:

• Functional linear models provide a powerful framework for genetic meta-analysis, accommodating various genetic variants (rare, common, or combined).
• The developed LRT and F-distributed statistics offer improved performance over existing methods like MetaSKAT.
• These novel methods hold promise for enhancing discoveries in whole-genome and whole-exome association studies.