Two things:
1) I wouldn't recommend using any structure with such a poor resolution
(i.e. restrict your data set to those structures that are refined, and
to a resolution of at least 2.5 angstroms or better)
2) 2TMA (Tropomyosin) is not a structure that should be included in an
assesment
of any method for prediction of globular protein secondary structures, since it
is not globular.
I would suggest that you have a look through your dataset and remove other
structures having these problems. Nobody would expect such things to
be included in a training/test set, since secondary structure definitions
are often unreliable, making assessment suspect.
Good luck.
Rob
Robert B. Russell Biomolecular Modelling, Imperial Cancer Research Fund
44 Lincoln's Inn Fields, P.O. Box 123, London, WC2A 3PX, U.K.
Tel: 44 171 269 3583 FAX: 44 171 269 3479 e-mail: russell at icrf.icnet.ukhttp://bonsai.lif.icnet.uk
