Fig. 6.

Neural Network. Selected features (here, seed region (SR) pairing, compensation (CP) pairing, thermodynamic (T) pairing, target site context (TSC) pairing and conservation (CV) pairing) are used as input signals in this feedforward partially connected neural network example. Each node decides what to send to the next node following various principles, such as fuzzy logic, genetic algorithms or Bayesian statistics. Weight factors are applied to each edge. Eventually, an output layer will combine all results in one or several nodes (one in this example), thus allowing the classifier to make a decision. The model can change the weights and node ordering to best classify the training data.