Secondary structure prediction

The question of how DNA sequence determines specific protein structure has been a constant source of fascination and speculation since the problem was identified. It remains an extremely difficult area; generally referred to as the ``folding problem'', it is one of the major outstanding questions in molecular biology. Many attempts have been made to predict the tertiary structure of a protein from its sequence. These fall into two distinct approaches:

• One approach is to set up a realistic mechanical model of the protein chain and simulate the folding process.
• Other approaches are empirical as they proceed by inference from known tertiary structures.

The approach to structure prediction based on mechanical models has the innate (possibly fatal) attraction that, in theory, it requires no prior knowledge of protein tertiary structure. If successful it could be applied uniformly to all sequences. By contrast, all methods based on inference from known structures are inherently limited in their applicability. They will only be appropriate for predicting structures similar to those which were used in the inference process. Fortunately there are often biophysical or biochemical clues that help make this decision and these are often integrated in the methods for structure prediction.

Currently the best way to achieve reasonable secondary structure predictions is to run a variety of prediction algorithms over your sequence and determine a consensus among the results. There are various web servers that will do these multiple analyses for you, including PIX at the HGMP and Jpred integrated in wEMBOSS at the Belgian EMBnet node:

http://www.hgmp.mrc.ac.uk/Registered/Webapp/pix/

http://www.be.embnet.org/wEMBOSS/

As yet, coverage of secondary structure prediction within EMBOSS is limited. More algorithms will be added to enable the conesensus approach described above. We'll take a look now at some of the predictions you can currently perform using EMBOSS.