For each gene tested in the collapsing analysis, we only retained the most significantly associated phenotype. Distinct gene–phenotype relationships from the collapsing analysis were partitioned into three categories (significant: P < 2 × 10−9 (binary n = 82, quantitative n = 269); suggestive: 2 × 10−9 < P < 1 × 10−7 (binary n = 113, quantitative n = 126); or non-significant: P > 1 × 10−7 (binary n = 18,551, quantitative n = 18,351)). The relationship between drug target status and gene–phenotype significance was assessed using Fisher’s exact test for each gene list. Specifically, for each of the five lists, we created a contingency table that included the number of significant collapsing analysis genes that intersected with the list and the number of genes that did not intersect with the list out of the list of genes tested in the PheWAS (n = 18,762). This was performed for both binary and quantitative traits. We also performed enrichment testing for OMIM32 genes and GWAS Catalog31 significant hits (both last accessed on 14 July 2020). We included the most significant associations per gene for the GWAS analysis.
