Evaluation of the relative performance of the subflattenings method for phylogenetic inference

TL;DR

This paper evaluates the effectiveness of subflattenings in phylogenetic inference, comparing them to flattenings through simulations, and finds they perform similarly but may be less biased, aiding future algorithm development.

Contribution

The study introduces software for constructing and evaluating subflattenings and provides a comparative analysis of their performance against flattenings in phylogenetic inference.

Findings

Subflattenings perform similarly to flattenings in split score distributions.

Subflattenings may be less affected by bias from split size and long branch attraction.

Simulations demonstrate the potential of subflattenings for improved phylogenetic inference.

Abstract

The algebraic properties of flattenings and subflattenings provide direct methods for identifying edges in the true phylogeny -- and by extension the complete tree -- using pattern counts from a sequence alignment. The relatively small number of possible internal edges among a set of taxa (compared to the number of binary trees) makes these methods attractive, however more could be done to evaluate their effectiveness for inferring phylogenetic trees. This is the case particularly for subflattenings, and our work makes progress in this area. We introduce software for constructing and evaluating subflattenings for splits, utilising a number of methods to make computing subflattenings more tractable. We then present the results of simulations we have performed in order to compare the effectiveness of subflattenings to that of flattenings in terms of split score distributions, and susceptibility to possible biases. We find that subflattenings perform similarly to flattenings in terms of the distribution of split scores on the trees we examined, but may be less affected by bias arising from both split size/balance and long branch attraction. These insights are useful for developing effective algorithms to utilise these tools for the purpose of inferring phylogenetic trees.