Airqtl's speed also enabled objective benchmarking and optimization of several methodological aspects:
✅ SceQTL P-value calibration for cell type specificity.
✅ Population admixture modeling. 11/n

When comparing GRNs between cell states:
🔹 Regulation directionality was largely conserved, suggesting chromatin-level rewiring.
🔹 GRNs were primarily determined by cell type, with cell condition as a secondary driver. 8/n

These de novo causal GRNs:
🔹 Quantify perturbation outcomes in primary human cell types.
🔹 Are not limited to transcription factors.
🔹 Align with published perturbation studies. 7/n

Its key is AIR, a novel data structure and algorithm that leverages the cell-donor hierarchy alongside an optimized implementation. See how AIR accelerates genotype-expression matrix multiplication and linear mixed model steps. 5/n

Check out airqtl: Our method is >10^8 faster than existing approaches. 🌪 4/n