Polygenic scores were constructed using cohort-specific, ancestry-stratified summary statistics for CAD and CAD-related traits, resulting in 51 GPS across all traits and ancestries. For each trait (for example, CAD) the best-performing combination of cohort-specific, ancestry-stratified GPSs was determined using stepAIC, and their optimal mixing weights ( β ) were determined using logistic regression in 116,649 individuals of European ancestry in the UK Biobank training dataset. The selected GPSs were linearly combined using these mixing weights to yield multi-ancestry scores predicting CAD for each trait (layer 1). The best-performing combination of multi-ancestry, trait-specific GPSs was determined using stepAIC, and their optimal mixing weights ( β ) were determined using logistic regression in 116,649 individuals of European ancestry in the UK Biobank training dataset. The selected GPSs were linearly combined using these mixing weights to yield GPS Mult (layer 2). Ancestries: AFR, African; EA, East Asian; EUR, European; HISP, Hispanic; SA, South Asian. Source GWAS traits: CAD 27 , 33 , 34 , 38 , 56 , body mass index (BMI) 38 , 57 , ischemic stroke 38 , 58 , 59 , diabetes mellitus (DM) 59 , 60 , 61 , peripheral artery disease (PAD) 38 , 56 , 62 , glomerular filtration rate (GFR) 38 , 63 , systolic blood pressure (SBP) 38 , 64 , diastolic blood pressure (DBP) 38 , 64 , LDL cholesterol 38 , 65 , 66 , HDL cholesterol 38 , 65 , 66 , triglycerides (TG) 38 , 65 , 66 .
