advertisement: compare things at compare-stuff.com!

Methods -- fold recognition using structure mapped library sequences

The method is a simple combination of the techniques previously described. Structural information is used to process the sequence information of library folds by means of self-organising maps. Unmapped query sequence property vectors (about which there should be no known structural information) are aligned against the mapped library sequence property vectors (as explained in Section 5.4.1 for hydrophobicity) exactly as described in the previous chapter in Section 4.2.

The mapping parameters were used as previously described. To summarise: training rate = 0.05, training cycles = 10000, number of trials = 20, map size factor = 1.0, map dimensions derived from the best-fitting ellipsoid. Except where otherwise stated, alignment gap opening and extension penalties are 3 and 0.5 respectively, and a local sequence window of 7 residues was used to calculate $S_{i,j}$.

The treatment of the hydrophobicity scales differs slightly from Section 5.4.1. Here the different scales have each been normalised by their mean and standard deviation for easy comparison. Therefore the weightings (or ratio to structure information) for the hydrophobicity information are not the same as those in Section 5.4.1 ( $f_{\overline{h}}$).