Multi-task learning for predicting synergistic drug combinations based on auto-encoding multi-relational graphs.

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Combinatorial drug therapy offers a promising strategy for treating complex diseases by exploiting synergistic drug effects.
• Predicting effective drug combinations remains a significant challenge due to intricate biological systems and incomplete knowledge of disease mechanisms and drug targets.

Purpose of the Study:

• To propose a computational framework, VGAETF (Variational Graph Autoencoder Tensor Decomposition), for predicting disease-related synergistic drug combinations.
• To leverage multi-relational graphs for modeling complex biological entity interactions.

Main Methods:

• Developed VGAETF, an end-to-end computational framework utilizing a multi-relational graph.
• Employed Variational Graph Autoencoder and Tensor Decomposition techniques to model complex relationships.
• Predicted synergistic drug combinations based on learned biological interactions.

Main Results:

• VGAETF demonstrated high predictive performance with an Area Under the Receiver Operating Characteristic curve (AUROC) of 0.9767 and an Area Under the Precision-Recall curve (AUPR) of 0.9660.
• The proposed framework outperformed existing methods in predicting synergistic drug combinations.
• Case studies validated VGAETF's effectiveness in identifying potential disease-related synergistic drug pairs.

Conclusions:

• VGAETF provides an effective computational approach for predicting synergistic drug combinations.
• The framework advances the application of graph-based deep learning in drug discovery and combinatorial therapy.
• VGAETF holds potential for accelerating the development of novel combination therapies for complex diseases.