Figure 2. Example of the impact of the branch length refinement. After the automatic parsing of the BLAST output, outlier sequences that have a "non common" distance with the query sequence are discarded. The distribution of the branch lengths appears more even in the recalculated phylogenetic tree (right). This refinement step improves the alignment quality. It also prevents artificially small sequence weights being obtained upon normalisation by the total branch length. The trees are calculated using the neighbor-joining procedure in ClustalW; they are then rooted using the query sequence as outgroup, as explained in the Sequence weights determination section.