PFRMAT SS 
TARGET T0065 
AUTHOR 5212-9839-7970 
REMARK 
REMARK 
METHOD RECURRENT BIDIRECTIONAL SECONDARY STRUCTURE PREDICTOR WITH 
METHOD INCORPORATION OF MULTIPLE EXTERNAL SECONDARY STRUCTURE  
METHOD RESOURCES 
METHOD 
METHOD 
METHOD Abstract: A prediction system based on neural network technology. 
METHOD   
METHOD The prediction system is trained on amino residues added with their corresponding secondary 
METHOD structure as an input array, which is matched with the target secondary structure during training. 
METHOD During prediction of each residue, the generated secondary structure is returned into the network 
METHOD by a feedback procedure.  
METHOD  
METHOD The prediction system performs an interactive data exchange between two neural network 
METHOD systems. Predictions are starting from the N-terminal side. After completion of the sequence, 
METHOD secondary structure data is transferred to the network system which itself is trained on sequences 
METHOD in reversed order. The reverse predictor continues with the reversed sequence and reversed 
METHOD feedback structure array at the C-terminal side. The completed structure array is transferred to 
METHOD the forward predictor and the prediction process continues and repeats until self-consistency is 
METHOD achieved. 
METHOD  
METHOD Different from Casp2, the initial assignment for secondary structure is built up using a weighted 
METHOD alignment of secondary structures combining 3 different structure resources: 
METHOD  
METHOD 1) A statistical predictor based on single sequence nearest neighbor algoritm with output filtering. 
METHOD Estimated Q3 accuracy is 60%. 
METHOD  
METHOD 2) A secondary structure threader in which secondary structure segments of the pdbselect* database  
METHOD are threaded along the target sequence using a scoring function based on PAM 250 matrix. Estimated  
METHOD accuracy on maximal scoring segments is 65%. 
METHOD  
METHOD 3) Blastp is used to align regions of the target sequence to fragments of a number of template  
METHOD proteins from the PDB database. The PDB coordinates of the fragments are translated into dssp  
METHOD secondary structure fragments. This method is applied only when sequence identities are greater then  
METHOD 25% and the fragment length of the aligned regions is greater then 30 residues. Gaps between aligned  
METHOD regions are left open.  
METHOD  
METHOD   
METHOD *june 1998 release of PDB_Select with 901 chains in the 25% sequence identity list 
METHOD  
METHOD  
MODEL  1 
M C  1.0 
K C  1.0 
N C  1.0 
A C  1.0 
K H  1.0 
Q H  1.0 
E H  1.0 
H H  1.0 
F H  1.0 
E H  1.0 
L H  1.0 
D H  1.0 
Q H  1.0 
E H  1.0 
W H  1.0 
V H  1.0 
E H  1.0 
L H  1.0 
M H  1.0 
V H  1.0 
E H  1.0 
A H  1.0 
K H  1.0 
E H  1.0 
A H  1.0 
N C  1.0 
I C  1.0 
S C  1.0 
P C  1.0 
E H  1.0 
E H  1.0 
I H  1.0 
R H  1.0 
K H  1.0 
Y H  1.0 
L H  1.0 
L H  1.0 
L H  1.0 
N H  1.0 
K C  1.0 
K C  1.0 
S C  1.0 
A C  1.0 
H C  1.0 
P C  1.0 
G C  1.0 
P C  1.0 
A C  1.0 
A C  1.0 
R C  1.0 
S C  1.0 
H C  1.0 
T C  1.0 
V C  1.0 
N C  1.0 
P C  1.0 
F C  1.0 
END 
