Extreme sampling design in genetic association mapping of quantitative trait loci using balanced and unbalanced case-control samples

It is extremely expensive to conduct large sample size array- or sequencing based genome scale association studies. For a quantitative trait, an extreme case-control study design may improve the power and reduce the cost of variant calling. We investigated the performance of extreme study design when various proportions of samples are selected from the tails of phenotype distribution. Using simulations, we show that when risk genotypes become rare in the population and effect size is relatively small, it is beneficial to carry out an extreme sampling study. In particular, the number of selected cases and controls can even be unbalanced such that power is further increased, compared with a balanced selection. Our application to two data sets: methadone dose data and yearling weight data, demonstrated that similar results for full data analysis can be obtained using extreme sampling with only a fraction of the data. Using power analysis with simulated data and an experimental data application, we conclude that when full data is unavailable due to restricted budget, it is rewarding to employ an extreme sampling design in the sense that there can be immense cost reductions and qualitatively similar power as in the full data analysis.