Minimum evolution using ordinary least-squares is less robust than neighbor-joining.

Area of Science:

• Computational Biology
• Phylogenetics
• Evolutionary Biology

Background:

• The minimum evolution method aims to reconstruct phylogenetic trees by minimizing the total branch length.
• Ordinary least-squares is commonly used to estimate tree length, with theoretical guarantees under certain conditions.
• Previous work suggested robustness when data is additive or close to additive on the true tree.

Purpose of the Study:

• To investigate the robustness of the minimum evolution method with ordinary least-squares for large numbers of taxa.
• To determine how deviations from additivity affect the accuracy of tree reconstruction.
• To compare the robustness of minimum evolution with neighbor-joining.

Main Methods:

• Theoretical analysis of the minimum evolution criterion using ordinary least-squares.
• Mathematical proofs examining the behavior of the difference in estimated tree lengths as the number of taxa (n) increases.
• Construction of specific examples demonstrating incorrect tree selection under slight data deviations.

Main Results:

• For large n, the difference between the estimated lengths of incorrect and correct trees can approach zero, even with additive data.
• A specific case is proven where minimum evolution incorrectly selects a tree W over the true tree T, with data deviating by only ~epsilon/4.
• This deviation threshold is half that required for incorrect selection by the neighbor-joining method.

Conclusions:

• The minimum evolution method, using ordinary least-squares, exhibits reduced robustness for phylogenetic reconstruction as the number of taxa grows.
• For large datasets, minimum evolution can be significantly less reliable than neighbor-joining due to its susceptibility to minor data inaccuracies.
• These findings highlight potential limitations of minimum evolution in complex evolutionary scenarios with many species.