A hierarchical clustering method for dimension reduction in joint analysis of multiple phenotypes

Area of Science:

• Genetics
• Biostatistics
• Computational Biology

Background:

• Genome-wide association studies (GWAS) identify genetic variants for complex diseases.
• Single-phenotype analysis in GWAS often reveals weak genetic associations.
• Joint analysis of multiple phenotypes offers greater power and biological insight.

Purpose of the Study:

• To develop a novel variable reduction method for joint analysis of multiple phenotypes in genetic association studies.
• To enhance the power of detecting genetic associations by reducing data dimensionality.

Main Methods:

• A two-step approach using hierarchical clustering method (HCM) for variable reduction.
• Step 1: Identify a representative phenotype for each cluster of correlated phenotypes.
• Step 2: Apply existing association testing methods to the representative phenotypes.

Main Results:

• Extensive simulations demonstrated that HCM significantly increases statistical power compared to analyses without HCM.
• HCM outperformed traditional methods like MANOVA, MultiPhen, and TATES in most simulated scenarios.
• The method's utility was validated using whole-genome genotyping data from a lung function study.

Conclusions:

• Hierarchical clustering method (HCM) is an effective strategy for reducing dimensionality in multi-phenotype association studies.
• HCM enhances the power to detect genetic associations with complex diseases by leveraging correlated phenotypes.
• This approach provides a more powerful framework for uncovering genetic underpinnings of complex traits.