The aim of Betaturns server is to predict different types of turns such
as Types I, II, IV, VIII and non-specific in a given amino acid sequence.
The method is based on neural network. It uses two feed-forward
back-propagation neural networks with a single hidden layer, where the
first sequence-to-structure network is trained on PSI-BLAST generated
position specific matrices (Altschul
et al. 1997). The second structure-to-structure network is trained on
the outputs obtained from first network and PSIPRED predicted secondary
structure information (Jones
1999). It has been trained and tested on a data set of 2881 sequence
unique proteins using five-fold cross-validation technique.
The results show that Type I and II beta-turns have better prediction
performance than type IV and VIII beta-turn types. The final network
yields an overall accuracy of 73.6%, 91.6%, 52.9% and 92.4% and MCC values
0.25, 0.36, 0.05 and 0.10 for Type I, II, IV and VIII beta-turns
respectively and is the highest achieved so far.
The input to the server is a single-letter code amino acid sequence in
fasta or free format. The server predicts the beta-turn types in two
steps. In first step, the residues forming beta-turns are predicted by
using BetaTPred2(Kaur
and Raghava 2003). The BetaTPred2 predicted beta-turns are further
classified into different types using betaturns server. The output
consists of target sequence, PSIPRED predicted secondary structure
(helix:'H', beta-sheet:'E' and coil:'C',). Turn residues are predicted as
4 residues block with turn types indicated by roman numerals I, II, IV,
VIII for turn types I, II, IV and VIII respectively or 'NS' for
non-specific beta-turn category which does not belong to any of the 4 turn
types.
