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

Population-scale single-cell RNA sequencing enables mapping genetic effects on gene regulation at the cellular level (for example, single-cell expression quantitative trait loci (sc-eQTLs)). However, existing sc-eQTL mapping approaches are either not designed for analyzing sparse counts in single-cell RNA sequencing data or become computationally intractable at scale. Here we developed jaxQTL-an efficient pseudobulk sc-eQTL mapping framework using highly efficient count-based models. In simulations, we demonstrated that jaxQTL with a negative binomial model provides calibrated inference while outperforming linear models. In OneK1K data, jaxQTL identified 16% more eGenes than linear models, while running substantially faster than single-cell Poisson mixed effect models. We found that fine-mapped sc-eQTLs are more distal from transcription starting site than those identified in all cells (bulk-eQTLs; P = 1 × 10^-4) and more enriched in cell-type-specific enhancers (P = 3 × 10^-10), revealing distal regulatory effects that are missed in bulk tissues. Although sc-eQTLs are largely shared across cell types, cell-type-specificity increases with distance to transcription starting site. Finally, sc-eQTLs explain a greater proportion of single nucleotide polymorphism-heritability than bulk-eQTLs across 16 blood and immune-related traits (9.90 ± 0.88% versus 6.10 ± 0.76%) and prioritize disease-associated genes missed by the bulk approach (such as IL6ST), improving, but not closing, the missing link between genome-wide association studies and eQTLs.