The predictive performance of polygenic scores (PGS) in clinical risk models is largely dependent on the number of samples available to train the score, together with the proportion of causal variants and heritability of the phenotype and and discovery-target cohort heterogeneity. Increasing the number of samples for a specific phenotype can be expensive and time consuming, but effective sample size can be increased by leveraging data from genetically correlated phenotypes. We propose a framework to generate enriched PGS from a wealth of publicly available genome-wide association studies, combining thousands of studies focused on many different phenotypes, into a multi-PGS. In the current study, the multi-PGS framework increased prediction accuracy over single PGS for all included psychiatric disorders and other available register-based phenotypes in a population-based case-cohort sample, with prediction R2 increases of up to 9-fold for attention-deficit/hyperactivity disorder (ADHD). We also generate multi-PGS for phenotypes without an existing PGS, like Asperger’s syndrome, with 9-fold increases in R2 prediction accuracy over the single autism spectrum disorder (ASD) PGS or for case-case predictions (e.g., ADHD vs. ASD), with the multi-PGS increasing 15-fold the R2 prediction accuracy over the single ADHD PGS. We benchmark the multi-PGS framework against other PGS re-weighting methods and highlight its potential application to new emerging biobanks or register-based genetic cohorts to generate PGS for every available phecode or defined phenotype in their system.