Rooting Phylogenetic Trees from Protein Alignments

A phylogenetic tree is a diagram that illustrates the relationships between species or organisms across time. Building phylogenetic trees is a crucial task in bioinformatics. Various approaches to perform this task such as parsimony methods, distance-based methods have been proposed, especially maximum likelihood methods. Normally, the maximum likelihood methods which based on time reversible substitution models can construct unrooted trees. To construct rooted trees, the maximum likelihood methods might use an outgroup or infer trees with the time non-reversible substitution models. Recently, the time non-reversible amino acid substitution models have been estimated for a number of clades that allow us to construct rooted trees from protein sequences with the time non-reversible models. In this paper, we investigated the performance of the maximum likelihood methods in reconstructing rooted trees by using the time non-reversible models and the outgroup technique based on real protein alignments. We used the Robinson-Foulds (RF) distance, the doublet distance and the triplet distance to compare reconstructed rooted trees with the true trees. Experiments showed that the outgroup technique helped build better rooted trees than the time non-reversible substitution model for small alignments. They performed equally well and could reconstruct the true rooted trees when using large alignments.