"Richard P. Grant" wrote:
> Please excuse the x-post. Which secondary structure prediction method
> is generally reckoned the 'best' these days. I know about
>http://www.EMBL-Heidelberg.DE/predictprotein/ would appreciate any
> other pointers.
see http://www.pasteur.fr/recherche/unites/neubiomol/ secstrpr.html
for pointers, references and a little discussion.
Important: use SEVERAL programs, based on DIFFERENT approaches
See the result of JPRED with PHD+DSC+NNSSP+PREDATOR
Currently the best are:
->statistical information (GOR, hydrophobicity etc ...)
DSC - It is not maintained anymore, because its son, PROF
is launched. However nothing is published. You can find DSC on the
pasteur website. Q3 ~ 70-72
->nearest-neighbor (~similarity with protein of know structures)
NNSSP Q3 ~ 68-72
->neural-networks
JNET
PSIPRED - it is a sort of reimplementation of PHD, with a
currated alignment as input, and a better training. Q3 up to 76%
->hmm
SAM99, aimed to be a fold-recognition program, can provide
good secondary structure predictions.
--
Dr Nicolas Le Novère e-mail: nl223 at cus.cam.ac.uk
Dpt Zoology, Univ Cambridge, Downing street, Cambridge CB2 3EJ, UK
tel: +44 1223 336623 fax: +44 1223 336676
