It is also important to filter SNPs based on minor allele frequency because statistical power is extremely low for rare SNPs. Figure 7 shows that the power to detect an association in a large dataset (n=10,000) with a relatively large effects (OR between 1.3 and 1.7) is extremely low for rare SNPs (<1% frequency).We recommend removing any extremely rare SNPs (including any monomorphic SNPs). The threshold chosen depends on the size of the study and the effect sizes expected. Power calculation software such as CaTS Power [43] or Quanto [44]can simplify power calculations for genetic association studies and inform the investigator of the allele frequency below which the study becomes severely underpowered. Minor allele frequency can be reported for each SNP using the --freq option in PLINK, and SNPs can be removed from the analysis using the --maf option, followed by a lower limit threshold. SNPs with frequency too low to yield reasonable statistical power (e.g. below 1%) may be removed from the analysis to lighten the computational and multiple testing correction burden. However, in studies with very large sample
