These databases are intendend to be used with hhblits. They were produced by a
(unfortunately) complex pipeline based on kclust with parameters set to 20%
sequence identity. The uniprot was first split into parts based on phylogenic
annotation to be conveniently parallel processed. The merged end results are
what most people will be interested in, but some might find the not merged ones
useful.

Andreas Hauser
andreas at creative-memory.de