A novel target ranking approach could be developed that relies on deep network representation learning of a PPI network annotated with disease-specific knowledge attributes derived from TCRD/Pharos. This involves mapping the enriched PPI network into a feature space using the neural network framework, Gat2Vec (45). Gat2Vec employs a shallow neural network model to facilitate joint learning on the structural and attribute contexts of a given network. In this case, the PPIs provide the structural context and the disease-related knowledge provides the attribute contexts. The feature space generated can be used to develop machine learning models that can predict the ranking of protein targets in the context of a disease. Additional protocols for extracting available data for specific targets of interest, as well the differences in knowledge availability between understudied and more studied targets, are discussed elsewhere (46).
