Comparisons were made with PHASE2 and fastPHASE5. PHASE is too slow for meaningful comparisons. For the shorter 15q25 region, fastPHASE and LRP processed 10,000 typed individuals in 92 hours and 16 minutes respectively. Based on the trio test, the discrepancy rate is 30.4% for fastPHASE and 0.066% for LRP. For individuals/SNPs it can phase, we estimate that LPR can phase a region 800 times longer than fastPHASE with a similar probability of not making any errors (Supplementary Results and Supplementary Table 2 online). These results reinforce the point that local phasing methods are not designed to phase long regions on their own. However, improvements to LRP could be achieved through incorporating local phasing ideas (Supplementary Note).
