OSDDlinux for Standalone, Galaxy & Local version
The CHpredict server predict two types of interactions: C-H...O and C-H...π interactions. For C-H...O interaction, the server predicts the residues whose backbone Cα atoms are involved in interaction with backbone oxygen atoms and for C-H...π interactions, it predicts the residues whose backbone Cα atoms are involved in interaction with π ring system of side chain aromatic moieties.

The method is based on the recurrent neural network (Jordan network) trained on single amino acid sequence and PSIPRED predicted secondary structure. It can predict those interactions where the donor and acceptor residues are separated by less than and equal to 16 residues. Two neural networks with a single hidden layer have been used where the first sequence-to-structure network is trained on single sequence encoded as binary bits. Further filtering has been done by using second structure-to-structure recurrent network trained on output of first net and PSIPRED predicted secondary structure. The training has been carried out using error backpropagation with a sum of square error function(SSE).



Please cite the following paper if you are using CHpredict for your research:

Kaur, H. and Raghava, G.P.S. (2006) Prediction of Cα-H...O and Cα-H...π interactions in proteins using recurrent neural network. In-Silico Biology 6, 0011