A new phylogenetic protocol: Dealing with model misspecification and confirmation bias in molecular phylogenetics

Lars S Jermiin; Renee A Catullo; Barbara R Holland

To illustrate the relevance and benefits of the fifth step in the new phylogenetic protocol, we examined the phylogenetic data used to infer the evolution of insects(3). The tetrahedral plots in Figure 3a -3c reveal that the nucleotide composition at the three codon positions is heterogeneous, implying that the evolutionary processes that operated at these positions are unlikely to have been time-reversible. However, the plots are deceptive because the presence of constant sites (i.e., sites with the same nucleotide or amino acid) in the data can mask how compositionally dissimilar the sequences actually are. To learn how to resolve this issue, it is necessary to focus on the evolution of two sequences on a tree (Fig. 3d ) and the corresponding divergence matrix at time 0 (Fig. 3e) and at time (Fig. 3f) . At time 0, the two sequences are beginning to diverge from one another, so the off-diagonal elements of the divergence matrix are all zero. Later, the divergence matrix may look like that in Figure 3f . All the off-diagonal elements are now greater than zero, and the so-called matching offdiagonal elements of the divergence matrix might differ (i.e., 23 ≠32 ). The degree of divergence between the two sequences can be inferred by comparing the off-diagonal elements to the diagonal elements, while the degree of difference between the two evolutionary processes can be inferred by comparing the above-diagonal elements to the below-diagonal elements. If the two evolutionary processes were the same, the matching off-diagonal elements in Figure 3f would be similar. A lack of symmetry (i.e., 23 ≠32 ) implies that the evolutionary processes along the two descendant lineages may be different. Because the dots in these plots do not fall along the diagonal line in the plots (showing that a lack of symmetry is not statistically significant), there is an overwhelming evidence that the evolutionary processes at these positions do not match a time-reversible Markovian process.