As introduced above, current PSs are often poorly transferable across populations. Considering PS as a normally distributed variable which is partially correlated with the trait of interest, we can mimic the poor transferability of PS across populations by applying a directional bias to the PS and reducing the correlation with the trait. In particular, this decrease in correlation and therefore predictivity can be absent, with fully transferable PSs, or complete, leading to a PS altogether disconnected with the trait of interest.
