A new quartet approach for reconstructing phylogenetic trees: quartet joining method

In this paper we introduce a new quartet-based method, called quartet joining, for phylogenetic inference. This method concentrates on reconstructing reliable phylogenetic trees while tolerating as many quartet errors as possible. This is achieved by carefully selecting two possible neighbor leaves to merge and assigning weights carefully to the quartets that contain newly merged leaves. Experiments on both simulated and real data show that our approach outperforms the quartet-based program TREE-PUZZLE and the distance-based method Neighbor-joining. Experiments on real data also shows the potential of this method. We also propose a simple technique to improve the quality of quartet set, which further improves the performance of the quartet joining.