Fig. 1

HEDDI-Net workflow: protein-drug affinity, disease similarity, feature embedding, and association learning model. A Protein-drug affinity model. This component predicts the binding affinities between various drugs and representative proteins. The shallow learning embedding processes the binding affinity data to generate a comprehensive affinity matrix for the drugs. B Disease similarity model. This part calculates the semantic similarities between diseases using the MeSH hierarchical structure. It generates a similarity matrix by embedding the graph-based similarities, which helps in identifying representative diseases. C Association learning model. This deep learning model integrates the embeddings derived from the protein-drug affinity and disease similarity models. It concatenates these features to predict the association probabilities between drugs and diseases using a multi-layer perceptron (MLP) with dropout layers to ensure robust predictions