Efficient count-based models improve power and robustness for large-scale single-cell eQTL mapping.

jaxQTL using a negative binomial count model yields calibrated sc-eQTL inference in simulations and identifies 16% more eGenes than linear models in OneK1K data.

• Why it matters: Existing sc-eQTL methods struggle with sparse single-cell counts or fail to scale, limiting reliable detection of regulatory variants at population scale.
• What they did: The framework performs efficient pseudobulk sc-eQTL mapping with highly optimized count-based modeling, then benchmarks it against linear models and single-cell Poisson mixed-effect approaches.
• The result: Fine-mapped sc-eQTLs are more distal from transcription starting sites (P=1×10^-4) and enriched in cell-type-specific enhancers (P=3×10^-10), and sc-eQTLs explain more SNP-heritability than bulk-eQTLs across 16 blood and immune traits (9.90±0.88% vs 6.10±0.76%).

AI summary of the abstract · may contain errors · not medical advice · read the paper.