Reconstructing Ancestor States with Maximum Likelihood: Support for One-and Two-Rate Models

摘要: Many evolutionary theories make predictions about the order and timing of character evolution, inferring ancestral states is considered a powerful tool for testing these (Maddison, 1994; 1995; Yang et al., Schluter 1997). The most common method inference maximum parsimony (MP; see Maddison Maddison, 1992, references therein). Although MP may offer reasonable estimates ancestors characters chosen to meet assumptions method, or closely related species, it perform poorly when much opportunity change exists (i.e., rate moderate high time has elapsed: Collins Omland, 1997; Zhang Nei, Yet, many traits interest are expected rapidly, new methods must be devised accommodate this. Such reconstruction now exist molecular sequences aminoacids (Yang Koshi andGoldstein, 1996; Felsenstein, pers. comm. [1995]) two-statediscrete continuous data (Martins, Schluter, 1997, Pagel,1999), but they have yet extensively assessed utilized (Cunningham, 1998). Most based on simple statistical models likelihood (ML) estimation techniques. Here, we consider one such (Schluter, 1997), which uses theMarkov-transition ratemodel infer changes in two-state discrete (Pagel, 1994). Because constrained single observation per tip (rather than DNA sites amino acids tip), behavior reconstructions not same as that genetic data. Crucial success an accurate estimate forward backward rates changebetween states.Becausemostdata sets toosmall warrant Žtting unequal rates, al. (1997) suggested accuracy best achieved by constraining equal. Using empirical examples, ask whether this was caution, can gained fromestimating two simultaneously. We Žrst brief review ML method. then present results from small survey published trees level support oneand two-rate trait evolution. Žnally introduce second sort model, differs between ingroup outgroup.