Fig. 1. A network can be represented by an adjacency matrix Y where each row and each column correspond to a specific node, with potentially different families of nodes associated with rows and columns. Each node is furthermore described by a feature vector, with potentially different features describing row and column nodes. For instance, row nodes nir can be proteins and column nodes njc can be drugs, with the adjacency matrix encoding drug–protein interactions. Proteins could be described by their PFAM domains and drugs by features encoding their chemical structure. Supervised network inference then consists of inferring missing entries in the adjacency matrix (question marks in gray) from known entries (in white) by exploiting node features.
