Genetic effects on gene expression often vary across cell types, motivating expression quantitative trait loci (eQTL) mapping in single-cell RNA sequencing data. Existing approaches generally aggregate expression across cells, use Gaussian models poorly suited to read counts or do not scale. Here we show that SAIGE-QTL, a Poisson mixed model that directly models single-cell read counts, enables calibrated and computationally scalable eQTL mapping while accounting for donor-level and cell-level covariates and unequal numbers of cells per donor. In simulations and three single-cell datasets, SAIGE-QTL improved statistical efficiency and discovery of genes with significant eQTLs (eGenes) relative to pseudobulk methods and produced results consistent with existing single-cell models at substantially lower computational cost. In 1.2 million immune cells, SAIGE-QTL identified 48.8% more eGenes than pseudobulk analysis. SAIGE-QTL further enabled genome-wide trans-eQTL mapping and rare variant set-based testing, identifying rare variant associations independent of common eQTL signals. SAIGE-QTL therefore expands single-cell eQTL discovery across common, rare and trans-acting genetic variation. SAIGE-QTL is a framework for eQTL mapping that directly models single-cell read counts using a Poisson mixed model, achieving improved statistical efficiency and power and enabling rare variant set-based testing in the single-cell eQTL setting.
