Fig. 1

Comparing MAFFT-linsi--add to UPP-add We ran MAFFT-linsi--add and UPP-add on a backbone alignment and two query sequences. The initial two sites in the MAFFT-linsi--add alignment each contains two letters from the query sequences, thus indicating it detects them as homologous to each other, although there are no letters from the backbone alignment in these two sites. This is impossible for any method that represents the backbone alignment by one or more HMMs and then adds sequences to the backbone alignment using the HMMs. Thus, UPP-add places these initial letters of the query sequences into separate sites. Visualization is done with WASABI [11]