In silico validation

We replicated the phenome-wide MR analysis for Category 4 phenotypes using different datasets for exposure (eQTL^56,57/pQTL^29,30) and outcomes. We developed three validation sets, with the first using a different outcome dataset (FinnGen⁵⁸). Validation 2 employed different exposure datasets (eQTL/pQTL) whereas Validation 3 utilized different exposure and outcome datasets from those used for discovery. Moreover, we used the MetaXcan framework to perform transcriptome-wide gene-based analysis that integrates large-scale transcriptome data (eQTLs) with the summary statistics of GWAS to validate the findings using a different methodology; we tested whether the predicted expression levels of the 30 prioritized genes were associated with dyslipidemia.⁴³ We explored potential causal genes in the prespecified five tissues (whole blood, subcutaneous adipose, visceral adipose, arterial, and liver tissues) that were selected based on LDSC heritability (Figure S2). The 30 therapeutic candidates from the five tissues were used to determine the correlation between predicted expression levels and dyslipidemia. Here, we defined p < 0.05 as the nominal statistical significance and set the Bonferroni correction threshold at <1.67E-3 (<0.05/30 therapeutic candidates). The targets were considered validated when they showed statistically significant association at least once in three MR validations or with MetaXcan for hypercholesterolemia or hyperlipidemia.