Figure 2.

Neural network design. The neural network consisted of three layers: input, hidden, and output. The inputs are the weighted fractions of each residue type and a conservation score for each position in a set of homologous sequences (see Methods). The total number of inputs was 189 because of 9 positions × 20 residues for each candidate, and nine conservation score inputs. Two sets of weight matrices were evolved through learning: one between the input and hidden layers and the other between the hidden and output layers. For the output layer, the corresponding expressions are highlighted below and have been left general so as to also apply to the hidden layer. I_i denotes input (i.e., output from hidden node i) and W_ji its corresponding weight. i denotes the hidden node number and j the output node number into which it goes. Y_j denotes the output from a given output node. Sigma is the threshold of a neuron. The expressions are similar for the hidden layer, with weights W_ji′ used in place of the previous weights.