CONSTRUCTING THE TREEFAM DATABASE

TreeFam is a database of phylogenetic trees of gene families. It aims to develop a curated resource that presents the accurate evolutionary history of all animal gene families, as well as reliable orthologs and paralog assignment. In developing TreeFam, four novel algorithms were designed to improve the accuracy of tree building or to serve special needs for development. The first is a constrained neighbour-joining that efficiently adds new sequences to an existing tree while maintaining the original topology at the same time. This method is used to expand a seed tree to a full tree without losing any information added by manual curation. The second algorithm is a leaf reordering that orders the leaves of a tree according to the weights of leaves. When it is drawn as a picture, one tree can be displayed in different ways, depending on the order of leaves. This algorithm helps to display trees in a consistent algorithm and facilitates visual examination of trees, which is particularly helpful when comparing two trees. Thirdly, duplication and loss inference is fit into a more general theoretical framework and extended to allow for a multifurcated species tree. A fourth algorithm has also been developed, which is a new algorithm for merging trees. The tree merge algorithm itself is not a tree building algorithm, but it reconstructs an optimal tree from several trees that are built from an identical sequence set with different tree building methods. The resultant tree should combine the advantages of, and so outperform, all the candidates. This is shown to occur successfully in a large-scale benchmark presented in the last chapter. This benchmark is one of the few evaluations that are based on real data in the phylogenetics literature. It also highlights the fact that each tree-building algorithm has its own strength, although ML and parsimonious methods are slightly better in general.