Chen et al. (2011) describe that the comparisons are based on thousands of blast scores, I presume it means by local alignments in very large numbers. This is supposed to be better for highly divergent proteins that might not align well by more popular tools like ClustalW.
Chen et al. (2011) Phylogenetic Characterization of Transport Protein Superfamilies: Superiority of SuperfamilyTree Programs over Those Based on Multiple Alignments. J Mol Microbiol Biotechnol 21:83–96 DOI: 10.1159/000334611