Null models are important for scoring and for doing local, rather than global alignment. You can see the local/global thing in @article{bucher-karplus, title="A Flexible Motif Search Technique based on Generalized Profiles", author="Philipp Bucher and Kevin Karplus and Nicolas Moeri and Kay Hoffman", journal="Computers and Chemistry", year="1996", month=Jan, volume="20", number="1", pages="3--24" } There are several papers on null models for scoring, many with Richard Hughey as one of the authors. I'm going to be resubmitting a revised paper on reverse-sequence null models soon. (The referees had some excellent suggestions that took a few months to implement and test, using student labor. It turned out that their ideas were wrong, but the testing needed to be done to show it.)