Hierarchical modeling in association studies of multiple phenotypes

Area of Science:

• Genetics
• Statistical genetics
• Bioinformatics

Background:

• Genetic studies often use disease-associated phenotypes due to ease of measurement and stronger genetic control.
• Rare phenotypes and multiple comparisons pose challenges in genetic association studies.
• Existing methods struggle with small sample sizes and numerous phenotypes.

Purpose of the Study:

• To develop and evaluate a hierarchical model (HM) for analyzing multiple disease-associated phenotypes.
• To improve accuracy of effect estimates and reduce false-positive rates in genetic association studies.
• To address limitations of conventional logistic regression models (LRM) for complex phenotypes.

Main Methods:

• Developed a semi-Bayes hierarchical model (HM) for multiple phenotypes.
• Utilized simulated data from the Genetic Analysis Workshop 14.
• Compared HM with conventional logistic regression models (LRM).
• Grouped 12 subclinical phenotypes into clusters: emotions, behavior, and anxiety.

Main Results:

• The hierarchical model (HM) slightly increased power for detecting true genetic associations.
• HM demonstrated a reduced false-positive rate compared to LRM.
• HM provided more accurate effect estimates for single-nucleotide polymorphism (SNP) associations with multiple phenotypes.

Conclusions:

• The hierarchical model (HM) is a powerful tool for genetic association studies involving multiple related phenotypes.
• HM effectively handles challenges posed by rare phenotypes and multiple comparisons.
• This approach enhances the reliability of genetic findings in complex disease research.