Neural Networks for Hierarchical Classification of G-Protein Coupled Receptors

G-Protein Coupled Receptors are an important protein family involved in signaling within a given cell. The functions performed by these proteins are organized in a hierarchy of classes, where each function corresponds to a class node. In this hierarchy, each class node can have one super-class and many sub-classes. We propose in this paper the use of Artificial Neural Networks to assign a given protein to a single class path of a hierarchy. This task is better known in the Machine Learning literature as hierarchical single-Label classification for protein function prediction. Our proposed method trains one neural network for each hierarchical level and combines the classes predicted in each level to provide the final classification. Experiments considering four different datasets in a comparison with other methods of the literature provided interesting results.