Genome-wide association studies have established many links between genes and disease but do not reveal the effect of most of the many possible variants within each disease-related gene. ...while the explosion in sequencing of human genomes has revealed millions of missense variants that change protein sequences, we only understand the phenotypic and clinical consequences of a minute fraction of these. Some methods can provide detailed mechanistic understanding, yet they can be time consuming since each variant is handled individually and further, they are most easily applied retrospectively. ...most current functional assays are challenging to scale to the almost 18,000 possible single amino acid substitutions in MSH2, making it difficult to assign pathogenicity to any new clinically discovered variant. ...each variant’s change in frequency is used to compute a score (normalised to wild-type fitness) that quantifies the effect of the variant on the property selected for. ...they measure mutation rates on a curated set of 185 variants from ClinVar and other clinical sources, which includes benign, pathogenic, and VUS, and find that the assay captures most of these pathogenicity classifications.