Mixture SNPs effect on phenotype in genome-wide association studies

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Traditional Genome-wide association studies (GWAS) face a "missing" heritability issue.
• Existing mixed linear models assume all single-nucleotide polymorphisms (SNPs) are associated with phenotypes.
• Most SNPs have minimal or no effect on phenotypes, necessitating more refined models.

Purpose of the Study:

• To propose an efficient Hierarchical Bayesian Model (HBM) for GWAS.
• To extend existing mixed models for automatic selection of significant SNPs.
• To address the common scenario where only a small subset of SNPs influences phenotypes.

Main Methods:

• Developed a Hierarchical Bayesian Model (HBM) using a mixture distribution for SNP effects.
• Employed Gibbs sampling for model estimation.
• Validated the HBM through simulation studies using Framingham Heart Study (FHS) data parameters and analyzed FHS and Health and Retirement Study (HRS) data.

Main Results:

• The HBM accurately estimates the proportion of associated SNPs and identifies significant SNPs in simulations.
• The HBM demonstrates adaptability to model mis-specification compared to standard mixed models.
• Analysis of FHS and HRS data identified specific SNPs (e.g., rs9939609, rs9939973 on FTO gene) associated with Body Mass Index (BMI), with findings replicated across studies.

Conclusions:

• The HBM provides a powerful approach for identifying significant genetic associations.
• The associated estimation algorithm effectively handles large SNP datasets common in modern genetics.
• The method successfully identifies SNPs linked to BMI and replicates findings in independent cohorts.