The Role of Subjectivity in Reconstructing Ancestral Character States: A Bayesian Approach to Unknown Rates, States, and Transformation Asymmetries

Abstract

The synthesis of data from ecology and systematics makes possible, at least in principle, the reconstruction of features of extinct ancestors. The reliability of such reconstructions, however, is a matter of considerable concern. For example, the commonly used method of parsimony character-state optimization (Farris, 1970; Hartigan, 1973; Swofford and Maddison, 1987) often (but not always) assigns unambiguous character states to internal (ancestral) nodes of fully resolved trees, but it is intuitively clear that some such assignments are more reliable than others (Fig. 1). Various probabilistic methods (including maximum-likelihood approaches) for determining the relative reliability of ancestral state assignments have been proposed, all depending on various model assumptions about character evolution (e.g., Frumhoff and Reeve, 1994; Maddison, 1995; Schluter, 1995; Schultz et al., 1996; Schluter et al., 1997; reviewed in Cunningham et al., 1998). These general models require plausible values for the parameters describing the probability of change between character states, and these values can either be estimated from the data or derived from independent sources. For example, maximum-likelihood reconstructions of nucleotide evolution (which are not necessarily explicitly concerned with ancestral states) usually derive theprobabilities of change between the four nucleotides from the relative base frequencies across all of thesequences under consideration (Felsenstein, 1981; reviewed in Swofford et al., A B A