Efficient Prioritization of Multiple Causal eQTL Variants via Sparse Polygenic Modeling.

Area of Science:

• Genetics
• Bioinformatics
• Genomics

Background:

• Traditional single-variant association analysis in expression quantitative trait loci (eQTL) studies struggles to distinguish causal variants, capture combined effects of multiple variants, and identify low-frequency variants.
• Sparse polygenic models offer a potential solution by simultaneously associating multiple genetic variants with gene expression, but their performance in eQTL studies remains unevaluated.

Purpose of the Study:

• To evaluate the efficacy of sparse polygenic models (Lasso, Elastic Net, BSLMM) in identifying causal eQTL variants.
• To compare the performance of these models against single-variant association analysis and a fine-mapping model using simulated and real data.

Main Methods:

• Utilized simulated genetic and gene expression data to assess the performance of Lasso, Elastic Net, and BSLMM.
• Compared sparse polygenic models against single-variant association analysis and a fine-mapping model.
• Analyzed RNA-sequencing and whole-genome sequencing data from 131 induced pluripotent stem cell (iPSC) samples.

Main Results:

• BSLMM significantly outperformed single-variant association analysis in prioritizing causal eQTL variants when multiple causal variants were present, showing 1.6- to 5.2-fold higher recall at 20% precision.
• BSLMM identified up to 2.3-fold more low-frequency variants as top eQTL SNPs compared to single-variant analysis.
• In real data analysis, BSLMM-identified eQTL SNPs exhibited higher functional enrichment in DNase hypersensitive (DHS) sites and were more frequently low-frequency.

Conclusions:

• Sparse polygenic models, especially BSLMM, are more effective than single-variant association analysis for identifying multiple causal eQTL variants at a single gene.
• BSLMM demonstrates superior performance in detecting low-frequency causal variants and prioritizing functionally relevant eQTLs.