MultiPhen: joint model of multiple phenotypes can increase discovery in GWAS

Area of Science:

• Genetics
• Statistical genetics
• Bioinformatics

Background:

• Genome-wide association studies (GWAS) typically analyze one phenotype at a time.
• This univariate approach may miss genetic variants associated with multiple related phenotypes.
• Existing multivariate methods can suffer from inflated type-1 error rates with non-normal data.

Purpose of the Study:

• To compare the performance of a novel multivariate GWAS method (MultiPhen) against the standard univariate approach.
• To introduce MultiPhen software for fast and interpretable simultaneous modeling of multiple phenotypes.
• To assess the power and accuracy of MultiPhen in detecting genetic associations.

Main Methods:

• Developed MultiPhen, a method using ordinal regression to model multiple phenotypes jointly.
• Tested MultiPhen via simulations to evaluate power and type-1 error rates.
• Applied MultiPhen to lipid traits in the Northern Finland Birth Cohort 1966 (NFBC1966) dataset.

Main Results:

• Simulations showed MultiPhen dramatically increases power for detecting variants affecting single or multiple phenotypes.
• MultiPhen demonstrated no inflation of type-1 error rates, unlike methods like CCA and MANOVA, with non-normal data.
• In real data, MultiPhen identified 21% more independent SNPs and provided a 37% increase in discovery when used alongside univariate GWAS.

Conclusions:

• MultiPhen offers a powerful and robust approach for multivariate genome-wide association studies.
• The method can uncover genetic associations missed by single-phenotype analyses.
• MultiPhen has potential applications in refining existing phenotype definitions and discovering novel heritable traits.