TABLE 6.

Performance of Evolved Neural Networks for the Discrimination of R5 and X4 Sequences Following Leave-One-Out Cross Validation in Terms of Area under the ROC Curve (A(z))

Number of Input Features	Number of Hidden Nodes	Number of Output Nodes	ROC curve area (A(z))
2	2	1	0.884
5	2	1	0.912
10	2	1	0.908
15	2	1	0.842
20	2	1	0.904
25	2	1	0.906
30	2	1	0.927

A maximum of 30 possible input features was available (2 domain charge features and 28 amino acid features). Forcing neural networks to use fewer than 30 inputs can still produce neural network models with reasonable A(z) values.