
PFRMAT SS
TARGET T0102
AUTHOR 9521-1365-5893
METHOD Program Name: PSIPRED
METHOD Synopsis: Neural network prediction based on PSIBLAST output
METHOD Five neural networks are used to predict secondary structure directly
METHOD from the profiles generated by PSIBLAST. Networks 1-4 predict
METHOD secondary structure, the 5th network filters the output of the first
METHOD level networks. The first level networks have 17*21 inputs and 75 hidden
METHOD units, the second level network has 17*3 inputs and 55 hidden units.
METHOD The networks have been trained on a set of 2937 proteins with
METHOD early-stopped training. 10% of the training data was used to detect
METHOD convergence.
METHOD The target sequence was scanned against a non-redundant databank of
METHOD filtered protein sequences (> 500,000 sequences), with 3 PSIBLAST
METHOD iterations.
METHOD NOTE: Single sequence weights used.
MODEL  1
M C 0.98
A C 0.59
K C 0.07
E C 0.18
F C 0.17
G C 0.82
I C 0.76
P C 0.28
A H 0.44
A H 0.49
V H 0.62
A H 0.63
G H 0.67
T H 0.63
V H 0.49
L H 0.59
N H 0.64
V H 0.67
V H 0.61
E H 0.36
A C 0.30
G C 0.56
G C 0.38
W H 0.45
V H 0.86
T H 0.87
T H 0.88
I H 0.94
V H 0.95
S H 0.97
I H 0.96
L H 0.97
T H 0.89
A H 0.80
V H 0.70
G C 0.43
S C 0.40
G C 0.39
G C 0.60
L H 0.48
S H 0.73
L H 0.65
L H 0.62
A H 0.64
A H 0.63
A H 0.33
G C 0.59
R H 0.77
E H 0.86
S H 0.85
I H 0.93
K H 0.96
A H 0.96
Y H 0.95
L H 0.96
K H 0.97
K H 0.95
E H 0.86
I H 0.91
K H 0.77
K H 0.25
K C 0.44
G C 0.87
K C 0.69
R E 0.35
A E 0.33
V E 0.46
I E 0.23
A C 0.47
W C 0.96
END

