FM-GPT: Bayesian fine mapping for phenome-wide transcriptome-wide association studies

Area of Science:

• Genetics and Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Transcriptome-wide association studies (TWAS) integrate genetic and gene expression data to link genes with traits.
• Spurious TWAS signals arise from linkage disequilibrium and correlated gene expression, necessitating fine-mapping methods.
• Large-scale phenomic resources enable multi-trait analyses, moving beyond single-trait genetic studies.

Purpose of the Study:

• Introduce FM-GPT, a Bayesian fine-mapping method for prioritizing causal genes across multiple correlated phenotypes in phenome-wide TWAS.
• Enable analysis of mixed outcome types (binary, count, continuous) within phenome-wide association studies.
• Reveal pleiotropic or phenotype-specific gene effects and disentangle complex gene-phenotype relationships.

Main Methods:

• Developed FM-GPT, a novel Bayesian fine-mapping approach for phenome-wide TWAS.
• Implemented gene-guided dimension reduction for phenotypes.
• Validated through simulations and applied to UK Biobank data for brain imaging and electronic health records.

Main Results:

• FM-GPT demonstrated superior accuracy in identifying true causal genes and controlling false positives in simulations.
• Applied to UK Biobank data, FM-GPT significantly reduced the number of candidate causal genes.
• Identified genes with pleiotropic effects on cortical thickness and genes influencing multiple medical conditions, revealing immune-metabolic trade-offs.

Conclusions:

• FM-GPT effectively prioritizes causal genes in large-scale phenome-wide studies with complex, multi-trait phenotypes.
• The method uncovers shared biological mechanisms underlying diverse human traits and diseases.
• FM-GPT advances translational research and the study of disease comorbidities.