I have a dataset, which includes ~100 clusters of orthologs from protist A and three distant relative taxa. I'd like to investigate which clade is closest to A protein for each cluster. First pBLAST hit definitely point to one clade in most cases, but on phylogenetic ML trees most A proteins cluster with another one (def. conditions, bootstraps are ok, multiple alignment done using MUSCLE). These two taxa have unequal coherence, so the only guess is that PhyML just groups all long branches (organism A + quite dissimilar organisms from second clade) together. On the other hand, PhyML results make much more biological sence, so I'm looking for a method to deal with this contradiction and find true closest relative.