PFRMAT SS
TARGET T0052
AUTHOR 1751-3146-3362
METHOD Using neural net fssp-1234-5-IDaa13-9-6-11-9-3-8-7-ehl-seeded-trained.net
METHOD This is a 4-layer network, with amino acid frequencies, insertions,
METHOD and deletions as inputs, and the following layers:
METHOD	window	units
METHOD	    9	    6
METHOD	   11	    9
METHOD	    3	    8
METHOD	    7	    3 (EHC)
METHOD 
METHOD The input amino acid frequencies were determined from weighted counts
METHOD and the recode2.20comp Dirichlet mixture regularizer.
METHOD The input alignment (using the SAM/T98 method) had only one sequence,
METHOD so we expect this prediction to be quite poor.  This network
METHOD was trained on multiple alignments (part of the target98 library),
METHOD and only gets about Q3=65% on single sequences.  The strand prediction
METHOD is expected to be slightly better then the helix prediction.
METHOD 
MODEL 1
L C 0.68
G C 0.72
K C 0.57
F C 0.46
S C 0.36
Q H 0.45
T H 0.43
C C 0.38
Y C 0.46
N C 0.45
S H 0.46
A H 0.41
I C 0.48
Q C 0.57
G C 0.63
S C 0.55
V E 0.54
L E 0.56
T E 0.48
S C 0.49
T C 0.58
C C 0.53
E C 0.47
R C 0.53
T C 0.73
N C 0.87
G C 0.91
G C 0.88
Y C 0.87
N C 0.84
T C 0.79
S C 0.73
S C 0.63
I C 0.59
D C 0.64
L C 0.61
N C 0.57
S H 0.54
V H 0.48
I C 0.47
E C 0.53
N C 0.75
V C 0.86
D C 0.84
G C 0.76
S C 0.65
L E 0.49
K E 0.58
W E 0.51
Q C 0.62
P C 0.75
S C 0.73
N C 0.52
F H 0.45
I H 0.52
E H 0.56
T H 0.44
C C 0.63
R C 0.68
N C 0.68
T C 0.67
N C 0.58
L C 0.51
A C 0.65
G C 0.80
S C 0.72
S H 0.71
E H 0.77
L H 0.84
A H 0.93
A H 0.95
E H 0.92
C H 0.83
K H 0.75
T H 0.79
R H 0.78
A H 0.73
Q H 0.67
Q H 0.52
F E 0.42
V E 0.44
S C 0.45
T C 0.44
K E 0.45
I C 0.54
N C 0.77
L C 0.79
D C 0.67
D H 0.51
H H 0.48
I C 0.47
A C 0.47
N C 0.59
I C 0.67
D C 0.74
G C 0.75
T C 0.74
L C 0.56
K C 0.46
Y C 0.44
E C 0.49
END