Jnet: A Neural Network Protein Secondary Structure Prediction Method

How and Why?

Jnet is a neural network prediction algorithm that works by applying multiple sequence alignments, alongside PSIBLAST and HMM profiles. Consensus techniques are applied that predict the final secondary structure more accurately. It was written as part of a continuing study to improve protein secondary structure prediction.

Jnet can also predict 2 state solvent exposure at 25, 5 and 0% relativeexposure. Positions where the different prediction methods do not agree are marked as no jury positions.  A separate network is applied for these positions, which improves the cross-validated accuracy. A reliability index indicates which residues are predicted with a high confidence.

Accuracy?

All predictions were tested on a set of 406 non-redundant proteins that were not used to develop the method. Residues predicted with a confidence value of 5 or greater, have an average Q3 accuracy of 84%, and cover 68% of residues. Relative solvent accessibility based on a two state model, for 25, 5 and 0% accessibility is predicted at 76.2, 79.8 and 86.6% respectively.

As a guide the cross-validated accuracy for Jnet at each stage is:

---------------------------------------------------------
Multiple sequence alignment only:          71.6%
Including the HMMer profile:               74.4%
Including the PSIBLAST frequency profiles: 75.2%
---------------------------------------------------------
Jury Decision if all files are available:  76.4% (8.4 sd)
---------------------------------------------------------

A paper has been published describing the exact details of the prediction method:
Cuff, J. A. and Barton, G. J. (1999) Application of enhanced multiple sequence alignment profiles to improve protein secondary structure prediction, Proteins 40:502-511 [Medline][Preprints]