Whole genome genotyping is now becoming feasible, under which the entire narrow-sense heritability would be represented. Assuming this is equivalent to about one million independent common SNPs [32], the required sample sizes are shown in Figure 4 and Table 5. Again, unless the heritability is explained by about 1000 markers, several tens to hundreds of thousands of subjects are needed to obtain a clinically useful AUC; for the genetic predictor to approach its potential, the order of magnitude is of the millions. The sample sizes to achieve AUC of 0.75 are larger than for 100,000 SNPs explaining half the heritability, but the latter scenario cannot achieve AUC of 0.99, so the clinical context can influence the choice of marker panel used to derive the predictor. It is clear that at current sample sizes, polygenic scores are only going to approach useful levels of discrimination if the marker panels include a high proportion of associated loci and the number of such loci is relatively small. Furthermore, for highly polygenic conditions the sample sizes needed to approach this potential are an order of magnitude higher than are currently available.
