Georg, I have forwarded your request to the genome browser people at UCSC, since it seems like their sort of problem. If I were faced with the problem, I would probably do a two-step search: 1) prefilter with blast using a very high E-value (like 1000) (probably taking the union of searches with each of the known examples) 2) run the output of the prefilter through SAM's hmmscore, with an HMM built from the known examples, and with the dbsize set to the number of sequences in the original database. The genome browser folks may very well have a better answer for you. Kevin Karplus