Significance Thresholds for Rare Variant Signals

With the advent of large-scale DNA sequencing studies, it is worth considering how to estimate the required significance threshold for tests of association between the resulting genetic variation and a phenotype of interest. Due to the rarity of most of the identified variants, standard analytic practice now includes, in addition to single-variant tests, a new set of statistical tests that consider simultaneously all genetic variability in a small chosen region of the genome. However, the question of how to set appropriate genome-wide significance thresholds for these region-based tests has received little consideration. To control the family-wise error rate, estimates of the effective number of independent tests, me, are required. Although for single-variant tests, me depends primarily on the linkage disequilibrium, for region-based tests, the choice of regions, of weights, and of test statistics will also influence me. Therefore, me will need to be estimated for each analytic plan. In this chapter, we review a recently proposed method for using the patterns of correlation between test statistics to estimate the required significance thresholds. In this approach, extrapolation from small sections of the genome to the whole genome can provide computationally feasible estimators for genome-wide significance thresholds. We also discuss other factors that may need consideration, such as exome sequencing, the use of false discovery rates for controlling type 1 errors, and region definitions that are not based on physical proximity.