Artificial Neural Network for the Prediction of Tyrosine-Based Sorting Signal Recognition by Adaptor Complexes
Figure 2
Neural networks for the analysis of Y-signals. (a) ANN architecture: the neurons in the network are represented by circles and the connections between units by arrows. The input layer is made up of 5 clusters (one for each X position within the XXXYXXØ motif) containing 20 nodes each (representing the 20 possible residues—only 3 per cluster is shown) plus the Ø cluster with only 5 nodes (for F, M, I, L, and V), yielding 105 neurons in total. Neurons from the hidden layer are labeled h1 and h2, whereas the output neuron is marked o, both types of units rely on a logistic activation function, depicted as a sigmoidal output-input response. Final network output is denoted as (Interaction Value). The weights associated to the input hidden layer and hidden-output layer connections are indicated as and , respectively. The Bias neurons are not showed. (b) ANN Signal analysis. Sequences (a hypothetical RSDYEPL signal is shown in red) are analyzed at every position. Within each of the 6 input clusters, only the neuron representing the aminoacid present at that position is activated (represented in red). This group of input neurons “fire” to the each hidden neuron according to the corresponding connection weight. Each hidden neuron compiles a total input and elaborates a sigmoidal output that is sent to the o-neuron, which in turn produces the network output .
