Fig. 3: Regulatory effects in Perturb-seq explain genetic association signals.
From: Causal modelling of gene effects from regulators to programs to traits
a, Gene effects on MCH can be predicted by regulatory effects on HBA1. Genes perturbed in Perturb-seq experiment are ordered by their effect sizes on MCH from LoF burden test. Perturb-seq β refers to log fold change of HBA1 expression after knockdown of the genes. Significant (P < 0.05) regulatory associations in Perturb-seq are connected with arrows. The protein structure of haemoglobin is presented using UCSF ChimeraX66 based on Protein Data Bank entry 1A3N. The P value is from the linear regression and is two-sided. KDreg, knockdown of a regulator. b, Enrichment analysis testing whether the top n HBA1 regulators (ranked by P values) are enriched at LoF or GWAS top hits. GWAS hits are the closest genes to the independently associated variants (Methods). Points indicate the odds ratio in the exact Fisher’s test. Enrichment was calculated with all the perturbed genes in Perturb-seq as a background. The error bars indicate 95% confidence intervals. The P values for the enrichment of the top 200 HBA1 regulators are 9.6 × 10−5 for the top 90 LoF hits, 0.65 for the top 90 GWAS hits and 0.01 for the top 543 GWAS hits. c, For every expressed gene in K562, regulator–burden correlation is plotted against their γ for MCH. The y axis shows the –log10(P) of the regulator–burden correlation, multiplied by the sign of the correlation. The P values are from the linear regression. Quadrants with a yellow background correspond to ‘concordant’ association, in which the sign of regulator–burden correlation aligns with the sign expected from the γ of the gene. d, Genome-wide QQ-plots for regulator–burden correlations among representative traits. Each dot represents one gene. Traits without significant signals lie along the dotted line. For other traits, see Extended Data Fig. 3c. P values are from the linear regression.
