Fig. 2: Airqtl offers eight orders of magnitude of acceleration and improved effect size estimation in sceQTL mapping. | Nature Communications

Fig. 2: Airqtl offers eight orders of magnitude of acceleration and improved effect size estimation in sceQTL mapping.

From: Airqtl dissects cell state-specific causal gene regulatory networks with efficient single-cell eQTL mapping

Fig. 2: Airqtl offers eight orders of magnitude of acceleration and improved effect size estimation in sceQTL mapping.The alternative text for this image may have been generated using AI.

a, b Running time (left Y) and computing cost (right Y) for sceQTL mapping on small-scale (a, defaults to \({n}_{{\mathsf{SNP}}}=500\), \({n}_{{\mathsf{gene}}}=500\), \({n}_{{\mathsf{donor}}}=100\), \({n}_{{\mathsf{cell}}}=1,000\)) and realistic large-scale (b) defaults to \({n}_{{\mathsf{SNP}}}=4,000,000\), \({n}_{{\mathsf{gene}}}=5,000\), \({n}_{{\mathsf{donor}}}=100\), \({n}_{{\mathsf{cell}}}=50,000\)) datasets with varying dimension sizes (X). c, d Ground-truth (X) and estimated (Y) sceQTL effect sizes (dots) by airqtl (c) and CellRegMap (d) without stratification. Deviation of the best-fit linear model (dotted line) from the diagonal (dashed line) indicates the extent of overall effect size underestimation. e, f Effect size estimation bias (e) and variance (f) for genes across expression quartiles (X). Dashed line: perfect performance. Color: method.

Back to article page