FIGURE 3.
The effect of ambiguous characters on topological support when both genes are evolving at the same rate. Axes and shades of gray are the same as in Figure 2. Note that the graphs in the left column of (a), (b), and (d) are identical to those presented in Figure 2. In an ML framework (a), ambiguous characters do not lead to a systematic bias in topological support, regardless of the rate of evolution (increasing from left to right columns). In a Bayesian framework (b), however, the magnitude and direction of the bias are a function of the rate of evolution. This bias is strongest when the rate of evolution is low or high and weakest when the rate of evolution is intermediate (e.g., when Gene A provides strong support for the true tree). When the rate of evolution is high, the bias exists when an exponential branch-length prior is assumed (b) but is absent when a uniform branch-length prior is assumed (c). The type of bias seen in the Bayesian framework can be demonstrated in the ML framework (d) if branch lengths are fixed at an arbitrarily low value (results for 10−6 substitutions per site per My are shown in the lower left graph) or a very high value (results for 1.0 substitutions per site per My shown in the lower right graph) data set. Note that in the Bayesian framework, the flat topological prior prohibits zero-length branches and the exponential branch-length prior penalizes long branches.