A method for protein secondary structure prediction
|1FXIA||78.12||83.33||72.92||76.04||77.08||58.33||81.25||3.12||96||Alpha and beta (a+b)||beta-Grasp|
This example shows an isolated prediction of chain A of Ferredoxin I, to give you a feel for how prediction and definition methods compare. The table above shows the Q3 accuracy for each of the methods. The quality value is a measure of where predictions go wrong in giving a value for where there are alpha helical predictions for regions that are actually beta sheet, and vice versa. This type of misprediction is much more damaging than, for example, predicting a helix that is too long, and will cause fold recognition methods to fail. The fold and class names are taken from SCOP.
The quoted accuracy for a method is always for an average performance over a number of proteins. This example should not be used as a performance measure, it merely shows you how the methods compare to a definition method. A more in depth measure of performance is given here. The definition methods used here are DSSP, STRIDE and DEFINE. Again, there is considerable variation within the different methods used for secondary structure definition.
In this example one can see that the 3 residue helix is mispredicted by nearly all the methods, (coloured green). This region of the structure is exactly where you find the 2Fe-2S cluster. If you have your MIME-types set to view Rasmol scripts you can use this to locate this area of structure within the protein. The commands,
'select FES', then 'cpk' typed in to the Rasmol window will highlight the 2Fe-2S clusters.
It is interesting though, that a 3% improvement in Q3 for this protein is seen if the 3 residue helix is not taken into consideration. This may state the obvious, but at least it gives you an idea as to how to interpret predictions, and their accuracy, especially as the STRIDE and DEFINE definition methods do not define this area of structure to be helical.
Barton group home page