To benchmark per-sample computational cost of reference-based phasing, we performed an additional set of analyses in which we phased 1,000 randomly selected samples (not contained in the simulated reference panels) in addition to the 72 trio children. We subtracted the Ntarget = 72 running times from the Ntarget = 1,072 running times to obtain the incremental cost of phasing 1,000 samples, thus adjusting for initialization costs (e.g., reading the reference data and synchronizing it with the target data), which account for a non-neglibible fraction of total computational cost when Ntarget is small. Finally, we divided by 1,000 to obtain per-sample costs and multiplied by 4 to scale up from the five chromosomes analyzed (≈25% of the genome) to a genome-wide analysis.
