To discover robust associations, replication in exome sequencing studies will be critical. Because small early studies will be inevitably underpowered, no gene may achieve exome-wide statistical significance. In such cases, unless strict correction for multiple tests is performed, researchers should resist the temptation to apply a battery of statistical tests, each with various weighing schemes and variant selection. We strongly argue that an association can only be considered real if it has been replicated. A reasonable replication strategy is to select a few genes (e.g., 10), based on the strength of association86 from the discovery stage and prior biological plausibility. Sequencing and rare-variant associations must then be performed on new samples, using multiple-test correction threshold applied only to the (smaller) set of candidate genes.
