Fig. 2

Overview of the UPP algorithm. The input is a set of aligned sequences. This sequence dataset is split into two parts: the backbone dataset and the set of query sequences. An alignment and tree are estimated for the backbone dataset, and an ensemble of HMMs is constructed based on the backbone alignment and tree. The query sequences are then aligned to each HMM, and the best scoring HMM for each sequence is used to add the query sequence to the backbone alignment. See text for more details