GUESS

GUESS implements a fully Bayesian variable selection approach for single- and multi-phenotype genome-wide association studies (GWAS) to identify SNPs and their contributions to complex trait combinations.

Key Features:

• Bayesian Variable Selection: Implements a fully Bayesian variable selection strategy to explore complex genetic-association models.
• Single- and Multi-Phenotype Analysis: Supports analysis of single and multiple responses in GWAS to detect SNP-trait associations in the presence of linkage disequilibrium and trait correlations.
• Hierarchical Prior Structure: Incorporates a flexible hierarchical prior for genetic effects that adapts to the correlation structure of predictors.
• Computational Implementation: Provides a computationally optimized C++ implementation with parallel processing capabilities via Graphics Processing Units (GPUs).
• SNP-specific Contribution Estimation: Identifies the specific contribution of each SNP to different combinations of traits.
• Support for Diverse Genomic Data: Applicable to diverse genomic data types including gene expression and exome sequencing.

Scientific Applications:

• Multi-phenotype GWAS: Applied to multi-phenotype genome-wide association studies to dissect shared and distinct genetic effects across traits.
• Refinement of Genetic Associations: Refines genetic associations and enhances interpretation of complex SNP-trait relationships.
• Gutenberg Health Study (GHS) Findings: In the GHS, identified associations such as SORT1 with TG-APOB and LIPC with TG-HDL that were not detected in larger meta-GWAS.
• Analysis of Molecular Data: Used for analysis of gene expression and exome sequencing data to link variants to molecular traits.
• Simulation Validation: Demonstrated increased power over both Bayesian and non-Bayesian multi-phenotype approaches through simulation studies.

Methodology:

Computational methods include a fully Bayesian variable selection approach with a hierarchical prior structure adaptable to predictor correlation, a computationally optimized C++ implementation, integration of GPU parallel processing, and estimation of SNP-specific contributions to trait combinations.