Practical phylogenetic usage of theoretical advances in distance-based tree learning

Many different statistical and computational tools for phylogeny inference are used in biology, but none currently take advantage of a body of theoretical work on fast-converging algorithms, which are designed to guarantee correctness with high probability even when sequence lengths are short relative to the number of taxa. Here, we provide a first implementation of one of the most advanced of these algorithms, and we assess its utility when applied in reasonable biological situations. Our simulation study shows that although the algorithm does report only correct relationships for short sequence lengths, it requires much longer sequences to produce well-resolved trees. We also find that realistic datasets will often not meet the assumptions of the algorithm, but that this largely does not compromise the correctness of the returned trees, though it can reduce their resolution. We additionally provide guidance on how the algorithm can be deployed when the true tree is not known, which is essential for any real-world application. Overall, our intention is to bring a class of algorithmic methods to the attention of the phylogenetics community, and to make the mathematical community aware of needs of practicing biologists.