Figure 1 shows the accuracy of imputation either using just the sequencing reads to impute genotypes or using the reads coupled with the 1000 Genomes Project reference panels2 (Online Methods). We observe high accuracies at ultra low-coverage (0.1–0.5x) when reference panels are used (Figure 1, Supplementary Note, Supplementary Figure 1). Sequencing at 0.2x coverage yields more than 90% of the effective sample size than is achieved by Illumina Human-1M-duo array plus conventional imputation, as assessed by average r2 to SNPs in the 1000 Genomes Project dataset for both common (>5% minor allele frequency) as well as low-frequency variants (1 to 5% minor allele frequency) (Figure 1). These simulations results suggest that sequencing at 0.1–0.5x coverage with imputation using the 1000 Genomes Project datasets can in principle achieve power comparable to high-density SNP arrays. These simulation results are robust to model assumptions and parameter values (Supplementary Note, Supplementary Tables 1,2,3).
