Fig. 7.

Phasing and imputation runtimes for the simulated reference panel [Table 1 (C)] containing 1 million samples (2 000 000 haplotypes) and the real-world COVID.Italy dataset [2113 samples; Table 2 (11)]. The numbers in brackets give the acceleration factor of EagleImp (phasing + imputation) compared to the corresponding Eagle2 (phasing only) run. Note that we could not perform imputation with PBWT using this panel due to insufficient main memory to generate the required .pbwt files, so only Eagle2 phasing times are shown. For $K=\text{max}$Eagle2 could not complete the analysis in any configuration due to insufficient main memory again (marked with ‘failed’), hence, no runtime or speedup could be measured in this configuration