2. Bioinformatics prediction of 2D structure

Although less popular than they once were, protein secondary folding prediction methods are proving very useful in guiding more elaborate structural model constructions. As with all prediction methods, the general principle is:

• assume that there is a relationship between the nature of the amino acid and its secondary structure and ;

• that can be identified using statistical data mining methods.

This requires a training set, i.e. a set of known structures for which the secondary structure has been assigned using a tool such as DSSP. The performance of the method is then evaluated on another set of structures, called a "set" or "validation set". Most secondary structure prediction tools are limited to a three-state prediction: helix, loop, strand. Learning methods range from...