Information Technology Reference
In-Depth Information
Table 3.
Results of a training session from which the methods PREDATOR2 and
PHD were excluded. (test set consisted of 2000 proteins)
number
name
weight
1
ChouFasman
3.0
2
PREDATOR2
-
3
Garnier
1.1
4
Simpa96
10.0
5
GOR4
6.8
6
DSC
6.2
7
DSC-l
5.6
8
CFpred
1.5
9
PHD
-
10
NNpredict
2.3
Table 4.
Comparing the weights assigned without PREDATOR 2 and PHD (column
A) or with both methods (column B). The last column shows the ratio of columns A
and B.
number
name
A(no
PHD/Pr.)
B(all
methods)
ratio(A/B)
1
ChouFasman
3.0
0.8
3.8
3
Garnier
1.1
0.3
3.7
4
Simpa96
10.0
1.1
9.1
5
GOR4
6.8
0.7
9.7
6
DSC
6.2
0.7
8.9
7
DSC-l
5.6
0.6
9.3
8
CFpred
1.5
0.3
5
10
NNpredict
2.3
0.5
4.6
The weight assigned to the PREDATOR 2 program drops dramatically in
comparison with its former weight when the second smaller dataset is used in the training
of the neural network. This is because the program PREDATOR 2 uses a database, which
contains sequences with "prototype" secondary structure predictions. When the sequences
to be predicted are not "known" by this database it can be seen in the results of the
secondary structure predictions accordingly.
Also the question remains if the information needed to improve the overall
prediction rate is present in the present secondary prediction methods. If the results of the
predicting programs are not complementary it is improbable to find improvement with
consensus or hybrid methods. Possible causes of the problems in protein secondary
structure prediction could be long range interactions and Cys-Cys disulfide bridges.
Another problem is the definition of secondary structure from 3D co-ordinates. This
definition is not exact because different algorithms to determine secondary structure are
used. For instance, DSSP and Stride agree in 96% of all residues [35], which leaves 4% of
the residue assignments open for interpretation.
The results of this work suggest that better methods should be used to construct a
consensus method that outperforms the best algorithm in the selection (PHD). For a
