Rare-variant tests can benefit from stratifying or weighting rare alleles by functional significance, as evidenced by simulations and sequencing studies of candidate genes61,64, 71–73. The power of rare-variant tests is strongly influenced by the fraction of causal variants among all variants analyzed and using functional information is an effective way to give greater weight to likely causal variants. For example, nonsense variants should be prioritized higher than nonconserved missense variants. Similarly, missense variants should be prioritized higher than synonymous variants. Functional consequences of variants can be predicted by examining effects of amino-acid changes using comparative sequence analysis and protein structure analysis. Many computational prediction and conservation74, 75 methods are available (reviewed in refs.76–79). The accuracy of those methods is around 80%80 and it is likely highest for rare variants (truly functional variants are most likely deleterious and kept at low frequencies by purifying selection, and so common variants are most likely neutral and nonfunctional). Therefore, using prediction methods enriches for functional variants and thus boosts the power of association tests. Because such predictions are not perfect, however, they should be
