At first glance it may appear that supermatrix methods are superior to supertree methods; however, both methods make a series of simplifying assumptions at different stages of data compilation and data processing that makes it difficult to judge which, if either, is actually superior. Supermatrix methods have the charm of being amenable to statistical analysis, something that is currently underdeveloped in supertree methods, but even a statistically significant result can be wrong if systematic errors are not eliminated.
One way to understand the impact of simplifying assumptions on the resulting tree is to use simulations, where the truth is known ([10
] and references therein). Unfortunately, simulations can cover only a tiny, tiny fraction of the universe of possible evolutionary scenarios. Thus, they only allow us to exclude phylogeny reconstruction approaches that already fail to show good performance for the selected simulation conditions. However, the converse is not true. The good performers under selected conditions are not necessarily good performers under all conditions. That is why all simulations have only a limited explanatory power. While the study by Kupczok and colleagues [10
] shows that supermatrix methods usually have a higher probability of inferring the truth, MRP-supertree methods are runners-up and are superior to supermatrix approaches in the case of significant disagreement between gene trees and the species tree. One should also note that the simulations refer to a sequence-based approach. It is at present unclear how to include morphological characters, due to the lack of generally accepted models. Thus, supermatrix approaches may be favored in the simulations.
Thus, for the time being, whenever one wants to study the evolutionary relationships of living organisms it is possibly best to apply many reconstruction methods and to discuss the differences and commonalities of the resulting trees. Only then one can distinguish between reconstruction artifact and true evolutionary history. Nyakatura and Bininda-Emonds [1
] discuss the outcome of different phylogeny reconstruction methods to avoid inferring wrong phylogenetic relationships. The ever increasing production of new sequence data and our increased ability to deal with complex models of sequence evolution will certainly lead to a further revision of the carnivore tree in ten years. However, the phylogeny presented today will help to understand where information is missing that needs to be filled in during the coming years.
Scientific progress draws upon the application of different methodologies to the same problem. Only conflicts in the results will lead to progress in our understanding of phylogenies and the relationship among organisms. We understand a lot about how phylogenetic inferences work, but our understanding of how tree inference from patchy data works is still in its infancy. The simulation studies published are too simplistic to come to sound conclusions. Thus, supertree methods carefully applied are still valid and relevant for phylogenetic inference.