A modified N-to-1 Neural Network, as used in the template-based experiments in this article. The left side is essentially the same as the standard N-to-1 Neural Network depicted in Figure 4. A further, single neural network maps homology information into a number of descriptors that are concatenated to the descriptors obtained from the sequence. In this case the feature vector f contains two parts, one sequence-based, one homology-based. The top feature-to-output network is essentialy the same as the standard N-to-1 NN case.