Polygenic scores (PGSs) are weighted sum scores of trait-associated alleles from up to millions of SNPs. As PGS research pivots to translation into health care settings a key issue for laboratories providing PGS is demonstration of analytical validity of PGS. We report data from 6 individuals who have been genotyped multiple times using the same and different technologies. These data were generated as part of standard experimental design for quality control purposes in two research settings over many studies and over many years. Using this opportunistic design of technical variability, we provide an empirical evaluation of technical reproducibility of PGS from 115 traits of different genetic architectures. Given a predefined set of SNP weights variability in PGS can reflect only SNP missingness or incorrect genotype call. We find very high reproducibility of SNP genotypes. In particular, the technical reproducibility of PGS generated from the same array technology and processed through the same quality control and imputation pipeline is very high. However, impact of missing SNPs varies between traits depending on the SNP’s weight for a trait. We provide a PGS quality score statistic (PGS:QS) that can be reported for each trait-specific score for an individual, to provide a quantitative assessment of the proportion of variation of the score that is captured by the SNPs genotyped/imputed for the individual. We provide an algorithm (PGS-impute) that updates the SNP weights of the scoring algorithm to the SNPs available for an individual, improving PGS accuracy. While validity of directly measured genotypes (whether from microarray or whole genome sequencing) is well-established, objective approaches to evaluate analytical reproducibility of PGS post-genotyping pipeline have been lacking. Here, we provide empirical data and an analysis framework which can be used by PGS providers to support understanding of analytical reproducibility and robustness.
Lin et al. (2026) studied this question.