What sample size would have sufficient power to detect association of the polygenic score? For breast cancer the heritability of liability is estimated as 44% [21]. If the marker panel explains half of this heritability, roughly as in the ISC study, then two samples each of 1978 cases and 1978 controls would have 80% power at nominal significance. For prostate cancer the heritability of liability is also 44% and 1766 cases and controls would be required in each sample. For the ISC study, assuming explained genetic variance of 28.7%, 735 cases and controls in each sample are sufficient. Thus it appears that association testing is well powered at current sample sizes if two independent studies are used for training and testing, but less well powered if a single sample is split into two subsets.
