Table 5 demonstrates the impact of our rewrite of –blocks. Due to a minor bug in PLINK 1.0’s handling of low-MAF variants, we pruned each dataset to contain only variants with MAF ≥0.05 before running –blocks. 95506 markers remained in the “synth1” dataset, and 554549 markers remained in “chr1”. A question mark indicates that the extrapolated runtime may not be valid since we suspect Haploview or PLINK 1.07 would have run out of memory before finishing.Table 5 –blocks runtimes (sec) ParametersDatasetMachineHaploview 4.2PLINK 1.07PLINK 1.90 –ld-window-kb 500 synth1Mac-2nomem3198.41.7Mac-12∼45 k3873.01.3Linux32-2nomem5441.13.4Linux64-512∼57 k2323.42.9Win32-2nomem9803.48.9Win64-2∼51 k5513.42.8 –ld-window-kb 1000 synth1Mac-2nomem6185.72.2Mac-12∼45 k7394.49.8Linux32-2nomem9876.810.0Linux64-512∼57 k4462.13.9Win32-2nomem18925.717.3Win64-2∼51 k10.3 k3.6 –ld-window-kb 500 chr1Mac-2nomem∼2700 k?550.9Mac-12nomem∼3600 k?426.0Linux32-2nomem∼4300 k?1288.4Linux64-512∼440 k?∼2600 k?1119.7Win32-2nomem∼17000 k?4535.8Win64-2nomem∼5700 k?1037.2This command breaks the genome into estimated “haplotype blocks” which are usually inherited together. The PLINK 1.9 implementation combines optimizations recently developed by Taliun et al. [26] with additional laziness and bit-level parallelism.
