Fig. 1: Overview of the cTWAS method.
a, Two scenarios that violate the assumptions of TWAS and lead to false-positive findings. X is a noncausal gene of the trait y. X is associated with the trait, shown as dashed arrows, because of the LD between its eQTL and nearby causal variants. Double-headed arrows represent LD between variants. b, The causal diagram implicitly assumed by TWAS. \(\tilde{X}\) represents the cis-genetic component of gene expression X. U represents an environmental confounder. c, The model of cTWAS. Gm, the genotype of mth variant; \(\tilde{X_{j}}\) and Xj, imputed expression and actual expression of the jth gene; θm, direct effect of the mth variant on the trait y; βj, effect size of the jth gene; ϵ, error term; δ0, point mass at 0. πG and πV are prior probability of being causal for genes and variants, respectively; \(\sigma^{2}_{G} \,\mathrm{and}\, \sigma^{2}_{V}\) are prior variance of the effect sizes of causal gene and variants, respectively. d, The workflow of the summary statistics version of cTWAS. The main steps of cTWAS are shown in the red boxes. cTWAS reports PIP for all genes and variants within LD blocks. P values for the genes and variants from marginal association tests are shown at the top as a comparison. Red dashed lines from the output panel indicate the genome-wide significance level or the PIP threshold for genes.
