On parsimony and clustering

Abstract

This work is motivated by applications of parsimonious cladograms for the purpose of analyzing non-biological data. Parsimonious cladograms were introduced as a means to help understanding the tree of life, and are now used in fields related to biological sciences at large, e.g., to analyze viruses or to predict the structure of proteins. We revisit parsimonious cladograms through the lens of clustering and compare cladograms optimized for parsimony with dendograms obtained from single linkage hierarchical clustering. We show that despite similarities in both approaches, there exist datasets whose clustering dendogram is incompatible with parsimony optimization. Furthermore, we provide numerical examples to compare via F-scores the clustering obtained through both parsimonious cladograms and single linkage hierarchical dendograms.