Analyzing metabolomics data for association with genotypes using two-component Gaussian mixture distributions.

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Standard methods for evaluating single nucleotide polymorphism (SNP) impact on quantitative phenotypes rely on linear models.
• These normal-based approaches may not be optimal for phenotypes better described by Gaussian mixture distributions, such as certain metabolomics data.

Purpose of the Study:

• To develop a novel statistical test for identifying SNPs affecting quantitative traits using Gaussian mixture models.
• To enhance the power and interpretability of genetic association studies for complex phenotypes.

Main Methods:

• Development of a likelihood ratio test (LRT) for mixing proportions of two-component Gaussian mixture distributions.
• Incorporation of biologically informed restrictions to further increase statistical power.
• Validation using simulated data and real-world data from the Framingham Heart Study (20,315 SNPs on chromosome 11).

Main Results:

• The proposed LRT and restricted LRT demonstrated improved power over standard linear and log-transformed linear models in simulations.
• Application to Framingham Heart Study data successfully identified known SNPs involved in fatty acid desaturation.
• The method showed enhanced sensitivity and interpretability compared to conventional approaches.

Conclusions:

• Gaussian mixture models provide a more suitable framework for analyzing certain quantitative phenotypes in genetic studies.
• The developed likelihood ratio test offers a powerful and interpretable approach for SNP association analysis, particularly for complex data distributions.
• This method advances the ability to detect genetic influences on quantitative traits, with implications for understanding complex diseases.