|Author||James A. Cuff, Geoffrey J. Barton|
|ML Method||Neural Network|
|Architecture||feed-forward back-propagation, single hidden layer|
Jpred itself is just a Protein Secondary Structure Prediction server, which exists since 1998 in different versions. JPred predicts Solvent Accessibility and Coiled-coil regions with the Lupas method and the secondary structure with the Jnet algorithm.
Jnet uses neural networks for its prediction. first, input: window of 17 residues over each amino acid in the alignment plus the addition of a conservation number; nine hidden nodes; three output nodes
second, input: window of 19 residues (result of first network) plus the conservation number; nine hidden nodes; three output nodes
input: multiple sequence alignment output: for each amino-acid three scores for three secondary structure elements (Helix, Sheet, Coil)