New global AllAllDB server at ETHZ

Gaston Gonnet gonnet at inf.ethz.ch
Sun Aug 22 13:45:41 EST 1993


In article <SRE.93Aug19170828 at al.cam.ac.uk>,
Sean Eddy <sre at al.cam.ac.uk> wrote:
>
>Heh. A good way to touch off a discussion is to say that something's
>"safely said" :)
>
>Do you mean "which can be detected by *pairwise* sequence alignment"?
>(I might even doubt that statement, depending on the scoring matrix
>you use, and how stringent a score of "80" (bits?) is.)  Profile or
==================================================================

We have used matrices which are derived in the same way as the
Dayhoff matrices were derived, but from a much larger sample.
Hence the score is -10*log10( Pr_of_such_alignment_happening ).

Hence a score of 80 indicates a prob of 10^(-8) (not too significant
when you think of the size of the database), a score of 200 indicates
a prob of 10^(-20), certainly more significant.


>hidden Markov model alignments, using statistical descriptions of a
>whole family, can potentially detect matches that pairwise alignments
>would consider insignificant.  Are you arguing that anything a profile
>or HMM can detect, there will be at least significant pairwise
>alignment in your database? I'd be tempted to bet a few beers against
>that proposition :)
======================================================
Why conjecture?  send a query involving your most contorted
example and see if it is found.

Best wishes, Gaston H. Gonnet, ETH Zurich.




More information about the Bio-soft mailing list