In our real data experiments we use haplotypes inferred by Merlin as the ‘true’ haplotypes for our methods comparison. In the Results section we show that SHAPEIT2 phases extended pedigrees with close to perfect concordance with the haplotypes produced by Merlin (typically average SE). This level of discordance is of a similar order to both the number of recombination events, and genotyping error [42] which the Lander-Green algorithm is known to be sensitive to. Whilst we have applied standard quality control procedures (including Merlin's error checking) to these data, genotyping errors are likely to still be present. Hence at least some of this discordance may be in fact due to errors in Merlin haplotypes. We also compare the recombination events detected by Merlin in extended families to those detected by our duoHMM approach. Any discordance between these crossover callsets may also be due (in part) to Merlin errors. We also wanted to investigate the ability of our method to call crossover events in duos and trios which cannot be done with the Lander-Green algorithm. For these reasons we created several simulated datasets to investigate these issues.
