On optimal gene-based analysis of genome scans

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Univariate marker analysis has limited power with multiple causal variants per gene.
• Gene-wide statistics combining variant effects are hypothesized to increase power.
• The performance and scalability of gene-wide methods for whole-genome scans, especially with sequence data, remain unclear.

Purpose of the Study:

• To assess the performance and scalability of various gene-based methods for genome-wide association studies (GWAS).
• To compare commonly used (VEGAS, GATES), less common (Simes, aSUM, kernel), and novel proposed methods.

Main Methods:

• Realistic simulations were employed to evaluate method performance.
• Methods assessed include VEGAS, GATES, Simes, adaptive sum (aSUM), kernel methods, and a proposed combined univariate/multivariate statistic.
• Scalability was considered for large-scale analyses like genome-wide sequence data.

Main Results:

• Simes demonstrated high speed and good power for single causal variant models.
• The aSUM method showed strong power for multiple causal variants, particularly with shorter genes.
• The proposed statistic exhibited good power across all tested causal models.
• A two-step strategy (Simes followed by aSUM or VEGAS) is recommended for sequencing data analysis.

Conclusions:

• A two-step genome scan analysis is recommended for sequencing studies, using Simes for initial screening and computationally intensive methods for refinement.
• The proposed method provides a viable alternative, offering good power with modest trade-offs.
• Method selection should consider gene length and the number of causal variants for optimal power in genome-wide association studies.