MultiPhen

MultiPhen performs joint genetic association testing by modeling linear combinations of multiple phenotypes with genotypes at single nucleotide polymorphisms (SNPs) to improve discovery in genome-wide association studies (GWAS).

Key Features:

• Joint Phenotype Modeling: MultiPhen employs ordinal regression to test the linear combination of phenotypes most associated with genotypes at each SNP.
• Increased Statistical Power: Simulation studies show MultiPhen increases statistical power relative to univariate GWAS, identifying variants that affect multiple phenotypes and those influencing only one phenotype.
• Robust Error Handling: MultiPhen maintains control of type-1 error across phenotype distributions and avoids inflation observed in methods that assume normal genotypes, such as canonical correlation analysis (CCA) and MANOVA, particularly for case-control or non-normal continuous phenotypes.
• Practical Applications: Applied to lipid traits from the Northern Finland Birth Cohort 1966 (NFBC1966), MultiPhen identified 21% more independent SNPs with known associations than univariate GWAS and, when combined with standard approaches, increased SNP discovery by 37%.
• Phenotype Refinement and Novelty: Linear combinations estimated at leading SNPs aligned with established formulas like the Friedewald Formula for lipid traits, facilitating refinement of phenotype definitions and the discovery of novel heritable phenotypes.

Scientific Applications:

• Enhanced GWAS Discovery: Detecting variants associated with multiple correlated phenotypes through joint testing to improve SNP discovery in GWAS.
• Analysis of Case-Control and Non-Normal Phenotypes: Providing association testing that controls type-1 error for case-control and non-normal continuous phenotypes compared with methods such as CCA and MANOVA.
• Lipid Trait Analysis and Phenotype Refinement: Applied to lipid traits (NFBC1966) to identify additional known SNPs and to derive linear phenotype combinations consistent with the Friedewald Formula.
• Novel Phenotype Identification: Revealing novel heritable phenotype combinations via linear combinations estimated at leading SNPs.

Methodology:

Ordinal regression to test linear combinations of phenotypes per SNP; simulation studies for power assessment; comparisons against canonical correlation analysis (CCA) and MANOVA for type-1 error evaluation; estimation of linear phenotype combinations at leading SNPs applied to NFBC1966 lipid traits.