We used the LD Score regression (LDSC [25]) python package to estimate the heritability explained by SNPs (SNP-h2). We used pre-computed LD scores (“eur_w_ld_chr/”), which are publicly available (https://data.broadinstitute.org/alkesgroup/LDSCORE/). LD scores were computed for every SNP using individuals from European ancestry from the 1000 Genomes Project. We restricted the analysis to well-imputed SNPs, filtered to HapMap3 SNPs, with MAF above 1%. We removed InDels, structural variants, strand-ambiguous SNPs and SNPs with extremely large effect sizes (χ2 > 80). Heritability was calculated on the liability scale by accounting for differences in population prevalence (4%) and sample prevalence (21%). The population prevalence was retrieved from the 2018 National Survey on Drug Use and Health [26].
