Bioinformatics: From Genome Sequences to Protein Structures

Structure prediction
Page 2 of 5

b. More advanced methods

The two methods we have used so far apply previously-computed statistical analyses to the sequence to be predicted, and tend to give fairly low prediction accuracies.

Accuracy can be improved by:-

  1. Obtaining a multiple-sequence alignment of all protein sequences related to the sequence in question. The highly conserved regions in the alignment tend to be those associated with regions of regular secondary structure (alpha-helices and beta-strands), whereas regions containing insertions and deletions tend to correspond to loops (random coil), as in the example below.
  2. Combining the predictions of several different prediction methods to obtain a consensus prediction.

    Here the bottom line gives a prediction based on some sort of weighted consensus among the six methods.

One site which makes use of both the above strategies is:

Jpred results

Try to run it, but do not use information from available structures - that would make it too easy.

Which fold class does the above prediction suggest?

How much of the protein is predicted as helix?

Now have a look at the line labelled "JNET Rel" in FULL HTML results. This gives relative level of confidence in its prediction. The green regions (associated with scores of 7-9 ) represent the highly-confident regions, most likely to have been correctly predicted.

How many regions of regular secondary structure might you be reasonably sure are correct?

Now let's see what we have if we take only the highly-confident regions. Paste the consensus prediction into the box below. Then paste the " PHD Rel" column below it. Then, below that, retype the consensus prediction so that it consists of dots, ".", at residue positions of low confidence, and the predicted secondary structure at positions of high confidence.

For example:- 

------EEEEEEEEEE-
97242258677767653
--.....E.EEE.E...
Confidently predicted regions

What percentage of the sequence has been confidently predicted?

Accessibility predictions

The one piece of information we haven't yet looked at are the accessibility predictions. These are in the line labelled "access" which predicts residues as being buried (B) or exposed (E) in the protein structure.

Of particular interest are alpha-helices or beta-strands which are either amphipathic or totally buried.

An amphipatic helix is one which, because it is on the surface of the protein, has one side consisting largely of hydrophobic residues which face the protein's hydrophobic core, and the opposite side consisting largely of polar residues which face out into the solvent.

As alpha-helices have a periodicity of 3.6 residues per turn, the pattern of buried (B) residues will be of the form: i, i+3, i+4, i+7.

Such a pattern would suggest a surface helix (as in the example shown, where i=1).

Similarly, an amphipatic strand has one side hydrophobic and the other polar. The geometry of strand residues means that the pattern of buried (B) and exposed (E) residues is a simple alternating one.

In this case the pattern suggests the strand is the "edge" strand of a beta-sheet, and pokes out into the solvent.

Beta-strands in alpha/beta proteins are often completely buried. These can be identified by a run of hydrophobic residues.

In the Jpred prediction this would show up as a run of residues classified as buried (B).

Can you identify any of the above features in the Jpred accessibility predictions for our mystery protein?

Carry on HERE

This material is prepared with the support of the project ESF pro V� II na UK, Reg. num.: CZ.02.2.69/0.0/0.0/18_056/0013322.