Biology Reference
In-Depth Information
Fig. 3.
Weight plots for ANN binary model of surface exposure (from
Holbrook (1990)).
139
(
A
) is the weight matrix for the buried residue predic-
tions; (
B
) is the matrix for the exposed residue predictions. For each window
position and amino acid type the weight of its connection (link) to the next
layer is represented as a shade of gray. Darker shades indicate higher weight
values. The amino acid residues are in order of decreasing hydrophobicity
according to Eisenberg scale.
140
were presented to the ANN as sliding windows of variable sizes (up to
13 residues) centered on the amino acid of interest. The analysis was
performed by scanning the entire sequence with these sliding win-
dows. To predict the surface exposure of the protein residues, the
investigators defined two categories for the buried (inaccessible) and
exposed (accessible) residues. Figure 3 shows a graphical representa-
tion of the weights in the ANN after training. Examination of the
