In a second set of analyses, we assessed imputation of newly discovered variants using as our test sets SNPs found in CEU and YRI by the complete ENCODE sequencing and CNPs. We created a reference panel of phased haplotypes that incorporated the new variant and the surrounding consensus genotype data, and used it to impute genotypes in additional samples. This models (for example) the imputation into an existing genome-wide association study of new SNPs and CNPs discovered by the 1000 Genomes Project or an exome sequencing project. We assessed imputation accuracy by masking each individual in the sample in turn and imputing its genotype from the rest of the sample, thus preserving the largest reference panel possible. For comparison, we also repeated the analysis by masking randomly selected, frequency-matched array SNPs, rather than newly discovered variants.
