Finally, we examined the variants in Table 1 by filtering them against known variation databases such as dbSNP, the 1000 Genomes Project variation data, or user-supplied list of variants. The ‘–filter’ argument in ANNOVAR is used for this purpose. The third variant in Table 1 (35delG in GJB2) is known to be an autosomal recessive mutation for deafness (21). Interestingly, it is annotated in dbSNP but not present in the 1000 Genomes Project. This example illustrates that filtering against dbSNP may sometimes fail to identify causal variants for Mendelian diseases, if the mutation is not private and has been well studied before. One additional useful feature of ANNOVAR is the ability to filter variants against pre-computed functional importance scores, such as SIFT scores (9), for all possible non-synonymous mutations in the human genome. For example, the R702W mutation in NOD2 was annotated as deleterious by SIFT (score = 0). Since pre-computed SIFT scores can be used, ANNOVAR is very efficient in the annotation, requiring several minutes to handle an exome using a modern desktop computer.
