MetaPhat

MetaPhat identifies which traits drive multivariate genotype-phenotype associations in genome-wide association studies (GWAS) of multiple correlated traits.

Key Features:

• Integration of Univariate GWAS Results: Leverages univariate GWAS results across multiple correlated traits to detect multivariate SNP associations.
• Decomposition of Multivariate Associations: Decomposes multivariate associations into central trait subsets using multivariate Canonical Correlation Analysis (CCA) with model selection based on Bayesian Information Criterion (BIC) and P-value statistics.
• Trace Plot Visualization: Produces trace plots of BIC and P-value statistics to visualize the contribution of each trait across model decompositions.
• Validation and Application: Validated on Global Lipids Genetics Consortium GWAS data and, in an application to 21 heritable and correlated polyunsaturated lipid species from 2,045 Finnish samples, identified seven independent loci associated with clusters of lipid species and reduced complex associations to three to five central traits.
• Technical Implementation: Implemented in Python with multi-processing support, quality control and clumping functionalities, and integrates with R for visualization.

Scientific Applications:

• Lipidomics: Identifies genetic loci and decomposes multivariate associations among correlated lipid species, including polyunsaturated lipids.
• Metabolomics: Dissects multivariate genotype associations across correlated metabolites to highlight key driving phenotypes.
• Complex trait studies: Improves specificity of phenotype–genotype associations in studies of correlated complex traits, aiding discovery of novel genetic loci and pathways.

Methodology:

Integrates univariate GWAS results, applies multivariate Canonical Correlation Analysis (CCA), uses Bayesian Information Criterion (BIC) and P-value statistics for model selection to decompose associations into central traits, generates trace plots, and supports multi-processing, quality control and clumping with R integration for visualization.