Surprising results on phylogenetic tree building methods based on molecular sequences
Department of Computer Science, ETH Zurich, Zurich, Switzerland
BMC Bioinformatics 2012, 13:148 doi:10.1186/1471-2105-13-148Published: 27 June 2012
We analyze phylogenetic tree building methods from molecular sequences (PTMS). These are methods which base their construction solely on sequences, coding DNA or amino acids.
Our first result is a statistically significant evaluation of 176 PTMSs done by comparing trees derived from 193138 orthologous groups of proteins using a new measure of quality between trees. This new measure, called the Intra measure, is very consistent between different groups of species and strong in the sense that it separates the methods with high confidence.
The second result is the comparison of the trees against trees derived from accepted taxonomies, the Taxon measure. We consider the NCBI taxonomic classification and their derived topologies as the most accepted biological consensus on phylogenies, which are also available in electronic form. The correlation between the two measures is remarkably high, which supports both measures simultaneously.
The big surprise of the evaluation is that the maximum likelihood methods do not score well, minimal evolution distance methods over MSA-induced alignments score consistently better. This comparison also allows us to rank different components of the tree building methods, like MSAs, substitution matrices, ML tree builders, distance methods, etc. It is also clear that there is a difference between Metazoa and the rest, which points out to evolution leaving different molecular traces. We also think that these measures of quality of trees will motivate the design of new PTMSs as it is now easier to evaluate them with certainty.