Decoding Genetic Markers of Multiple Phenotypic Layers Through Biologically Constrained Genome-To-Phenome Bayesian...

Area of Science:

• Genomics and Phenomics
• Statistical Genetics
• Computational Biology

Background:

• Multivariate analysis of genomics and phenomics data faces scalability and interpretability challenges.
• Existing methods struggle to integrate prior biological knowledge effectively.

Purpose of the Study:

• To introduce Bayesian Genome-to-Phenome Sparse Regression (G2PSR), a scalable and interpretable method for joint genomics and phenomics analysis.
• To overcome limitations in current genome-to-phenome association studies.

Main Methods:

• G2PSR utilizes a Bayesian neural network framework with sparse SNP-gene constraints.
• Incorporates prior knowledge of variant-gene associations and employs variational dropout for gene sparsity induction.
• Enables estimation of gene and SNP uncertainty in phenotypic data reconstruction.

Main Results:

• Demonstrated effectiveness on synthetic and real datasets, outperforming state-of-the-art methods.
• Successfully applied to an imaging-genetics analysis of Alzheimer's Disease Neuroimaging Initiative data.
• Achieved accurate selection of relevant genes in datasets with high SNP-to-sample ratios.

Conclusions:

• G2PSR offers a scalable and interpretable solution for genome-to-phenome association studies.
• The method enhances statistical power by preventing multiple testing correction and assessing combined SNP effects.
• G2PSR effectively identifies relevant genes, addressing key limitations of current approaches.