Source of the recursive Ncut–neighbor-joining performance difference

Identify whether the observed difference in phylogenetic reconstruction accuracy between recursive normalized-cut (Ncut) reconstruction and neighbor joining (NJ) arises from the affinity transformation of pairwise evolutionary distances, the recursive graph-partitioning procedure, or both.

Background

The study compared recursive Ncut reconstruction using affinity matrices derived from pairwise distances with NJ applied directly to the same distance matrices. WAG- and JC69-based distances produced higher split recovery with NJ, whereas BLAST-derived logarithmic distances favored recursive Ncut under some conditions.

The authors explicitly note that the experimental design evaluated complete reconstruction pathways and therefore could not disentangle the effects of converting distances into affinities from the effects of recursively partitioning the resulting graph. Determining the source of this performance difference would clarify which component of the reconstruction pipeline is responsible for the contrasting results.