We performed a confirmatory factor analysis using GenomicSEM34, which is robust to sample overlap and sample-size imbalances35. Within the EUR-like population we used the provided linkage disequilibrium (LD) scores36. For the AFR-like models, we used within-sample LD scores calculated using cov-LDSC37. We carried the best fitting models forward for multivariate GWAS. We ran the GWAS results from each population through FUMA (Functional Mapping and Annotation of GWAS)38 to identify independent loci (r2 < 0.1) among genome-wide significant SNPs. To further quantify the polygenic overlap between externalizing and suicide attempt, we applied bivariate MiXeR39 which uses a bivariate Gaussian mixture model to estimate the proportion of shared influential genetic variance between two traits. Finally, we ran all results through a standard series of post-GWAS pipelines (detailed in the supplementary information).
