Polygenic scores

Effect sizes per variant come from a genome-wide association study, that is a weight describing how strongly a risk allele shifts the value. The score multiplies, for each variant, the number of effect alleles carried by its weight and sums all contributions. The PGS Catalog provides these weight lists in a standardised, citable format so a computation stays reproducible and traceable. Each model carries a stable identifier such as PGS000004 that records the source study, effect type, and genome build.

Genome does not use an external score engine; it applies the PGS Catalog weight file directly through its own matching pass. It reads the score file line by line, recognises the columns for rsID, effect allele, and effect weight including the harmonised hm_ columns, and looks up each variant in the sample's genotypes. Per variant it counts the dosage of the effect allele, that is 0, 1, or 2 copies, multiplies it by the weight, and adds the contribution to a running sum; markers flagged as recessive count only at two copies. Before summing, Genome checks at the first shared rsID that the score and SNP files match the same reference genome, and aborts on a divergent position or chromosome. Optional HLA alleles in a model are resolved against an HLA*LA bestguess.txt instead of being estimated from SNP dosages.

A weighted sum score is only as complete as the markers actually found in the genome. Genome therefore reports openly how many of the markers listed in a model are covered and in how many the effect allele truly appears, and writes the PGS identifier and citation into the result. Missing markers silently set their contribution to zero, which can shift the absolute score downward; this is why the coverage figure is part of the finding and not just a footnote. For short-read WGS this transparency matters, because models can originate from microarray studies and individual positions may be absent depending on the calling pipeline and coverage.

A high score means elevated average risk in the reference population, not certainty for an individual. For populations underrepresented in the source studies, scores are often poorly calibrated, because allele frequencies and linkage patterns differ between ancestry groups; the same score can then land in the wrong part of the distribution. Genome reports only the raw weighted sum together with coverage and does not place it on a percentile or risk scale, since that would require a matching reference distribution. PGS thus remains a research quantity, not a clinical test.