Multivariate Generalized Linear Models for Twin and Family Data.

Area of Science:

• Quantitative Genetics
• Biostatistics
• Statistical Modeling

Background:

• Multivariate twin and family studies are crucial for understanding disease inheritance and gene-environment interactions.
• Current analyses often rely on structural equation or linear mixed models, primarily for Gaussian data.
• Extending these analyses to non-Gaussian data remains a challenge.

Purpose of the Study:

• To propose a unified statistical modeling framework for multivariate Gaussian and non-Gaussian twin and family data.
• To extend classical biometric indices, like heritability and correlations, to non-Gaussian datasets.
• To provide a flexible approach for modeling mean, variance, and covariance structures.

Main Methods:

• Developed a flexible and unified statistical modeling framework.
• Accounted for non-normality by modeling mean-variance relationships.
• Utilized a linear covariance model, allowing dispersion components to be functions of covariates.
• Employed marginal specification for model extension.

Main Results:

• The proposed framework successfully analyzes both Gaussian and non-Gaussian twin and family data.
• Extended classical biometric indices, including bivariate heritability and genetic, environmental, and phenotypic correlations, to non-Gaussian data.
• Demonstrated model utility through simulation studies and six real-world data analyses.

Conclusions:

• The proposed unified framework offers a flexible and powerful tool for multivariate genetic and environmental analyses.
• This approach enables more comprehensive insights into disease inheritance and gene-environment interplay across diverse data types.
• Computational implementation is available via the R package mglm4twin.