NEDD

NEDD predicts novel drug-disease associations by applying meta-path-based network embedding to undirected heterogeneous networks that integrate drug-drug similarity, disease-disease similarity, and known drug-disease associations to facilitate drug repositioning.

Key Features:

• Meta-path-based network embedding: Uses meta paths of varying lengths to capture indirect relationships and high-order proximities within a heterogeneous network.
• Heterogeneous network integration: Integrates drug-drug similarity, disease-disease similarity, and known drug-disease associations into an undirected heterogeneous network.
• Low-dimensional representations: Translates complex biological interactions into low-dimensional representation vectors for both drugs and diseases.
• Predictive modeling: Employs a random forest classifier to predict potential novel drug-disease associations from the learned embeddings.
• Benchmark validation: Validated on a gold standard dataset comprising 1,933 validated drug-disease associations and reported to outperform existing state-of-the-art methods.

Scientific Applications:

• Drug repositioning: Identifies new therapeutic uses for existing drugs by predicting novel drug-disease associations.
• Prediction of novel indications: Predicts candidate drug-disease pairs that can be pursued as new drug indications.
• Prioritization for experimental follow-up: Narrows the search space for experimental validation by computationally ranking likely associations.

Methodology:

Integrates drug-drug similarity, disease-disease similarity, and known drug-disease associations into an undirected heterogeneous network; employs meta paths of varying lengths to capture indirect and high-order proximities; converts network information into low-dimensional representation vectors for drugs and diseases; uses a random forest classifier to predict novel associations; validated on a gold standard dataset of 1,933 validated drug-disease associations.