Co-regulation and LD between genes can confound TWAS signals and lead to spurious associations for non-causal genes. Conditional analysis and finemapping was performed to distinguish genes with increased evidence of exerting an independent causal effect on AN. In other words, we tested whether there was statistical evidence to support each of our identified association signals as directly relevant for the disorder. Conditional analysis (Gusev et al., 2016) estimates the residual independent association of TWAS signals after controlling for the predicted expression of nearby significant genes. Benjamini-Hochberg significant TWAS genes were subjected to conditional analysis to predict which genes accounted for the localised signal. From 313 significant TWAS signals, 97 genes had a conditionally independent association (pJoint < 0.05) with AN as indicated by their nominally significant joint p value (online Supplementary Table S4). The gene most significantly associated with AN, WDR6 (Zconditional−TWAS = 7.4, p = 1 × 10−13, Cortex) maintained an independent association after conditioning on the 77 other TWAS significant gene models (21 unique genes) from chromosome 3, three of which are also conditionally independent (CTNNB1, GOLIM4, STX19).
