RLFDDA: a meta-path based graph representation learning model for drug-disease association prediction

Area of Science:

• Bioinformatics
• Computational Biology
• Drug Discovery

Background:

• Drug repositioning is crucial for identifying new drug efficacies.
• Predicting drug-disease associations (DDAs) computationally remains challenging.
• Effective integration of diverse biological features is key to improving DDA prediction accuracy.

Purpose of the Study:

• To propose a novel meta-path based graph representation learning model, RLFDDA, for predicting potential DDAs.
• To enhance the accuracy of discovering new indications for approved drugs.

Main Methods:

• RLFDDA calculates drug-drug and disease-disease similarities.
• A heterogeneous network is constructed integrating DDAs, disease-protein, and drug-protein associations.
• Meta-path random walks learn latent representations, concatenated for joint drug-disease association predictions, followed by random forest classification.

Main Results:

• RLFDDA demonstrated superior performance (AUC, F1-score) on benchmark datasets compared to state-of-the-art models.
• A case study on paclitaxel and lung tumors showed high validation rates for predicted associations.
• The model offers a promising approach for novel DDAs discovery.

Conclusions:

• RLFDDA effectively predicts drug-disease associations by integrating heterogeneous biological networks.
• The model's performance suggests a valuable new perspective for drug repositioning and novel indication discovery.
• RLFDDA provides a robust computational framework for advancing pharmaceutical research.