Kernelized approach enables explainable gene prioritizations for complex traits

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Genome-wide association studies (GWAS) identify genetic variants linked to traits, but pinpointing specific effector genes within loci is difficult.
• Existing methods like PoPS prioritize candidate genes using functional profiles but lack transparency in their predictions.
• The biological plausibility of single-gene nominations from GWAS loci often remains uncertain.

Purpose of the Study:

• Introduce Kernelized Polygenic Priority Score (K-PoPS), a novel method for interpretable effector gene prioritization in GWAS.
• Enable gene-centric explanations by decomposing predictions into contributions from training genes.
• Provide tools to assess the biological plausibility and support for nominated effector genes.

Main Methods:

• Reformulated the PoPS algorithm using kernelization to enable prediction decomposition.
• Developed an 'anchor score' to quantify support from user-defined trait-relevant genes.
• Implemented K-PoPS using a full-feature ordinary least squares (OLS) approach.

Main Results:

• K-PoPS improved closest-gene enrichment over default PoPS for 26 of 37 traits in the Pan-UK Biobank dataset.
• Predictions supported by anchor scores showed greater enrichment for closest-gene proxies compared to unsupported predictions.
• K-PoPS provided biologically plausible explanations for nominating *SCARB1* over *UBC* for apolipoprotein B levels and identified multiple plausible genes for dilated cardiomyopathy.

Conclusions:

• K-PoPS offers a valuable post hoc framework for interpreting and examining GWAS effector-gene nominations.
• The method enhances transparency and biological validation of gene prioritization in genetic association studies.
• K-PoPS facilitates a more nuanced understanding of complex genetic architectures within GWAS loci.