Fig. 3From: BioBLP: a modular framework for learning on multimodal biomedical knowledge graphsThe architecture of BioBLP for the BioKG graph. On the left we illustrate the graph with the three entity types that have attribute data: proteins (P), molecules (M, representing drugs), and diseases (D). We omit entity types without attributes for clarity. For the three types, we retrieve relevant attribute data, which are the inputs to encoders for each modality. The output of the encoders is a vector representation of the entitiesBack to article page