Reports on the lack of computational reproducibility of molecular phylogenies, produced in multiple runs of the same program or by different programs, cast a long shadow on downstream research using these phylogenies and on efforts to build the tree of life. We show that this irreproducibility does not decrease the accuracy of the reconstructed evolutionary relationships. Also, we find inferred molecular phylogenies to have log-likelihoods comparable to or better than the true phylogeny. Therefore, the lack of computational reproducibility of molecular phylogenies is not a problem for evolutionary studies.

The simulated datasets were generated by a previous study under an autocorrelated rate model (Tamura et al. 2012).