Polygenic Risk Scores (PRS) were generated using the PRS-cs software package (19). The PRS-cs auto approach calculates PRS by assuming a general distribution of effect sizes across the genome, and then reweighting Single Nucleotide Polymorphisms (SNPs) based on this assumption, their effect size in the original GWAS, and their linkage disequilibrium (LD) patterns from 1000 genomes phase 3 to create weights for every SNP that are then summed for a final score. We chose PRS-cs so that all common variation across the genome could be leveraged given the polygenicity of substance use phenotypes. Our models were trained using two well-powered GWAS: 1) Drinks per week (DPW, original GWAS N=537,349) from the GWAS & Sequencing Consortium of Alcohol and Nicotine use (GSCAN) (20), and 2) a problematic alcohol use (PAU) meta-analysis of GWASs (12)of: A) alcohol use disorder (specifically, alcohol dependence from the Psychiatric Genomics Consortium (11), B) ICD codes for alcohol dependence (ICD-9: 303.* or ICD-10: F10.2*) and alcohol abuse (ICD-9: 305.* or ICD-10: F10.1*) from the Million Veteran Program(12,21), and C) the problem subscale of the Alcohol Use Disorders
