3 <title>JNet Secondary Structure Prediction</title>
6 <strong>JNet Secondary Structure Prediction</strong>
7 <p>Secondary structure prediction methods attempts to infer the
8 likely secondary structure for a protein based on its amino acid
9 composition and similarity to sequences with known secondary structure.
10 The JNet method uses several different neural networks and decides on
11 the most likely prediction via a jury network. <br>
14 Cole C., Barber J.D. and Barton G.J. (2008) The Jpred 3 secondary structure prediction server
15 <em>Nucleic Acids Research</em> <strong>36</strong> W197-W201</li>
17 Cuff J. A and Barton G.J (1999) Application of enhanced
18 multiple sequence alignment profiles to improve protein secondary
19 structure prediction <em>Proteins</em> <strong>40</strong> 502-511</li>
22 The function available from the
23 <strong>Web Service→Secondary Structure
24 Prediction→JNet Secondary Structure Prediction</strong>
25 menu does two different kinds of prediction, dependent upon the
26 currently selected region:
29 <li>If nothing is selected, and the displayed sequences appear to
30 be aligned, then a JNet prediction will be run for the first sequence
31 in the alignment, using the current alignment. Otherwise the first
32 sequence will be submitted for prediction.</li>
33 <li>If just one sequence (or a region on one sequence) has been
34 selected, it will be submitted to the automatic JNet prediction server
35 for homolog detection and prediction.</li>
36 <li>If a set of sequences are selected, and they appear to be
37 aligned, then the alignment will be used for a Jnet prediction on the <strong>first</strong>
38 sequence selected in the set (that is, the one nearest the top of the
39 alignment window).</li>
41 <p><strong>Note</strong>: JNet secondary structure prediction is a
42 'non-column-separable' service - predictions are based on the sequence
43 profile of contiguous stretches of amino-acid sequence. A prediction
44 will only be made on the visible parts of a sequence (see <a
45 href="../features/hiddenRegions.html">hiding columns</a>) as if it were
46 a contiguous polypeptide chain. Prediction accuracy at the hidden column
47 boundaries may therefore be less than indicated by JNet's own
48 reliability score (see below).</p>
49 <p>The result of a JNet prediction for a sequence is a new annotated
51 <img src="jnetprediction.gif">
52 <p>The sequence for which the prediction was made is the first one
53 in the alignment. If a sequence based prediction was made then the
54 remaining sequences in the alignment are the aligned parts of homologs
55 which were used to construct a sequence profile for the prediction. If
56 the prediction was made using a multiple alignment, then the original
57 multiple alignment will be returned, annotated with the prediction.</p>
58 The annotation bars below the alignment are as follows:
61 <li>Lupas_21, Lupas_14, Lupas_28<br>
62 <em>Coiled-coil predictions for the sequence. These are binary
63 predictions for each location.</em></li>
64 <li>JNETSOL25,JNETSOL5,JNETSOL0<br>
65 <em>Solvent accessibility predictions - binary predictions of 25%,
66 5% or 0% solvent accessibility.</em></li>
68 <em>The consensus prediction - helices are marked as red tubes,
69 and sheets as dark green arrows.</em></li>
71 <em>The confidence estimate for the prediction. High values mean
72 high confidence. prediction - helices are marked as red tubes, and
73 sheets as dark green arrows.</em></li>
75 <em>Alignment based prediction - helices are marked as red tubes,
76 and sheets as dark green arrows.</em></li>
78 <em>HMM profile based prediction - helices are marked as red
79 tubes, and sheets as dark green arrows.</em></li>
81 <em>Jpred prediction - helices are marked as red tubes, and sheets
82 as dark green arrows.</em></li>
84 <em>PSSM based prediction - helices are marked as red tubes, and
85 sheets as dark green arrows.</em></li>
87 <em>Amino Acid frequency based prediction - helices are marked as
88 red tubes, and sheets as dark green arrows.</em></li>
90 <em>A '*' in this annotation indicates that the JNETJURY was
91 invoked to rationalise significantly different primary predictions.</em></li>