Analysis of multilocus models of association

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Common diseases result from multiple genetic variants interacting with each other and environmental factors.
• Complex genetic models require sophisticated statistical methods for accurate inference.
• Traditional methods like Bonferroni correction may lack power in detecting true genetic associations.

Purpose of the Study:

• To evaluate the effectiveness of controlling the false discovery rate (FDR) for statistical inference in complex multivariate genetic models.
• To compare FDR-based methods against traditional model selection procedures like forward stepwise regression.

Main Methods:

• Statistical inference based on controlling the false discovery rate (FDR).
• Simulation analysis comparing FDR methods with forward stepwise regression.
• Evaluation of false positive rates and statistical power.

Main Results:

• FDR-based methods maintain a false-positive rate close to the nominal rate.
• Forward stepwise regression selected numerous non-contributing genetic variants.
• FDR methods demonstrated better statistical power compared to Bonferroni correction when null hypotheses are false.

Conclusions:

• FDR-based methods provide a viable approach for evaluating complex, multivariate genetic models.
• These methods offer improved control over false positives and enhanced power over Bonferroni correction.
• The power of any method is ultimately limited by sample size and model complexity.