Fast and reliable reconstruction of phylogenetic trees with very short edges extended abstract

Phylogenetic reconstruction is the problem of reconstructing an evolutionary tree from sequences corresponding to leaves of that tree. A central goal in phylogenetic reconstruction is to be able to reconstruct the tree as accurately as possible from as short as possible input sequences. The sequence length required for correct topological reconstruction depends on certain properties of the tree, such as its depth and minimal edge-weight. Fast converging reconstruction algorithms are considered state-of the-art in this sense, as they require asymptotically minimal sequence length in order to guarantee (with high probability) correct topological reconstruction of the entire tree. However, when the original phylogenetic tree contains very short edges, this minimal sequence-length is still too long for practical purposes. Short edges are not only very hard to reconstruct; their presence may also prevent the correct reconstruction of long edges. In this paper we present a fast converging reconstruction algorithm which returns a partially resolved topology containing all edges of the original tree whose weight exceeds some (non-trivial) lower bound, which is determined by the input sequence length, as well as some properties of the tree, such as its depth. It does not depend, however, on the minimal edge-weight. This lower bound provides a partial reconstruction guarantee which is strictly stronger than the guarantees given by other fast converging algorithms. Our algorithm also has optimal complexity (linear space and quadratic-time) which, together with its partial reconstruction guarantee, makes it appealing for practical use.