[Dundee Uni] [Jpred]

The Barton Group

A method for protein secondary structure prediction


Documentation - Technical Detail

Input Protein sequence(s):
  • raw sequence
  • MSF alignment
  • BLC alignment
Output 3 state secondary structure prediction, in:
  • HTML
  • Postscript
  • PDF
  • Jalview (Java viewer and editor)
  • Original ASCII output
Prediction method
  • Jnet; two fully connected, 3 layer, neural networks, the first with a sliding window of 17 residues predicting the propensity of coil, helix or sheet at each position in a sequence. The second network receives this output and uses a sliding window of 19 residues to further refine the prediction at each position.
  • Coils
Alignments
  • PSI-BLAST scan of a Seg and helixfilt filtered SP/TRMBL database with a 3 iteration scan
  • Perform pair-wise comparison of all sequences, to generate percentage identities
  • Cluster sequences on basis of percentage identity (complete linkage)
  • Remove members with >75% identity
  • Remove gaps from the target sequence, and corresponding column beneath that gap, use this alignment for all methods
  • Extract PSI-BLAST and HMMPSSM profiles for Jnet
Predict
  • Display in coloured HTML, Java, Postscript and PDF
Consensus
  • Per residue combination of Jent prediction network