Sensivitity of the maximum parsimony algorithm to missing data

A phylogenetic algorithm computes a tree of distance relationships on a set, S, of phylogenetic descriptions (which may not be complete), given a phylogenetic-description transformation function, D, defined on S. Maximum Parsimony (MP) is a widely used phylogenetic algorithm that computes the shortest phylogenetic tree that represents the tree distances on S determined by D. To date, the sensitivity of MP to missing/incomplete data has not been systematically investigated. Although a general characterization of this sensitivity is intractable, robust empirical characterizations for typical MP configurations are possible. Here, I present an analysis of the sensitivity of several commonly used tree robustness metrics to missing/incomplete data, for a widely used MP implementation and typical MP problem set-up, applied to randomized mid-sized missing-data sets. The results show a counterintuitive limitation of one of those robustness metrics.