I tried to rely on strand annotation from dbEST, but soon realized it is often wrong. I think this is due to clones being inserted the wrong way into the vector, I guess the rate this happens depends on the kit being used. This also means that although the researcher "knows" the orientation, she will often be wrong.

Thanks for bringing this up. I have had really poor results from the various EST cleaning tools, some old notes at http://blog.malde.org/index.php/2008/05/08/cleaning-up-sequences/. Do you have any independent resources demonstrating the effectiveness of these tools?

If the question is: did I create an artificial "EST" set with vectors/ribosomal sequences thrown in, then point mutated/ mutated with indels/flipped (all this can be seen in real EST data), then looked at results, then answer is: not yet. I am mapping 3+ millions of various species ESTs to a novel genome, some part of EST sets I mapped without seqclean pre-processing, so I hope to see the difference. GMAP, which I use for ESTs mapping is often overeager to call a match (on a protein level) with UTRs, and possibly with other sequences, so less dubious ends should give less nonsense matches.