Question about freak structures used in testing of prediction algorithm

Lemieux Sebastien lemieuse at ERE.UMontreal.CA
Thu Jul 20 13:07:29 EST 1995


STRELETS at SCRI.FSU.EDU ("VICTOR B. STRELETS") writes:

>I'm finishing paper about prediction of secondary structure and find
>some freak case(s) of accuracy (less than 20% per-residual in the
>simulation of blind test). There are only two such structures,
>both are tropomyosins (2TMAA and 2TMAB). They contain only one
>secondary structure element: alpha helix of length more than
>250 amino acids. It seems reasonable from the point of view of
>their function, but I never thought before that it can be obstacle 
>when used as a prediction input.

  Most of the published secondary structure prediction algos are tuned
to predict structure of "globular proteins" and tropomyosine is not
one of that kind!

  Before testing an algo, you should take care about what are the
limitation of the method and apply it when choosing your sample from
the databank.

Good luck!


--
| Sebastien Lemieux, dept. biol.  ||  Look behind the wave of changes
| lemieuse at alize.ERE.UMontreal.CA ||  Feel the future taking shape
| PGP public key on finger.       ||  I can see the world to come (KJ)
  http://alize.ere.umontreal.ca/~lemieuse/



More information about the Molmodel mailing list