FIGURE 2.

The algorithmic bias of PhyML and RAxML. ML trees were built by PhyML (left panel) and RAxML (right panel) for (A) identical sequences, (B) saturated sequences, and (C) sequences generated from a 4-taxon star tree (${\text{S}}_1:0.01$, ${\text{S}}_2:0.01$, ${\text{S}}_3:0.01$, ${\text{S}}_4:0.01$). The proportions of three ML trees $\left(\left(S_1,S_2\right),S_3,S_4\right)$, $\left(\left(S_1,S_3\right),S_2,S_4\right)$, and $\left(\left(S_1,S_4\right),S_2,S_3\right)$ were calculated and compared with the expected proportions $\left(1/3,1/3,1/3\right)$ if two phylogenetic programs PhyML and RAxML do not have an algorithmic bias.