in their filtering procedure (3). We confirmed that it was predicted as benign by SIFT as well (SIFT score = 0.18). We next filtered the variants from the 1000 Genomes project and dbSNP version 130, assuming that variants observed in public databases are less likely to be causal variants for Miller syndrome. This logic is similar to that used in two exome-sequencing studies (3,11), although they did not utilize 1000 Genomes Project data sets. This process left us with 413 variants. Next, in the reduced variants set, we assessed whether multiple rare variants exist in the same gene as compound heterozygotes. Interestingly, only 23 genes were left by this analysis. Finally, we assessed whether some of these 23 genes belong to a group of ‘dispensable’ genes, that is, genes with high-frequency nonsense mutations (in >1% subjects) in the 1000 Genomes Project. The underlying rationale is that such genes are unlikely to be causal for a very rare Mendelian disease, or that such genes are highly susceptible to sequencing and alignment errors in short-read sequencing platforms. Three genes (ZNF717, FAT1, OR4C3) were deleted, and we were left with 20 candidate genes, including the causal gene DHODH. Each of the methods described
