We focus on PS of four thoroughly studied traits (Type 2 Diabetes (T2D)28, height29, Body Mass Index (BMI)29 and breast cancer30. By introducing the concept of partial PS and applying it on the sample groups above, we find that a small portion of the genome is enough to improve trait predictions and that such approach can be used to correct for population level PS bias. Finally, we test the predictivity of ancestry specific partial PSs and their combination on datasets for which both phenotypic and genotypic information are available, namely the Estonian Biobank31 and the UK Biobank27. The results show that, when GWAS data are available for more than one ancestry, the combination of multiple partial PSs improves trait predictability in individuals with a mixed genetic background.
