Deep graph embedding for prioritizing synergistic anticancer drug combinations.

Area of Science:

• Computational biology
• Pharmacology
• Artificial intelligence in medicine

Background:

• Drug combinations enhance cancer treatment efficacy and overcome resistance.
• Experimental screening of all possible drug combinations is costly and time-consuming.
• Integrating multiple networks for synergistic drug combination prediction using deep learning is underexplored.

Purpose of the Study:

• To propose a Graph Convolutional Network (GCN) model for predicting synergistic drug combinations specific to cancer cell lines.
• To leverage multimodal network integration for improved prediction accuracy.

Main Methods:

• Developed a GCN model employing heterogeneous graph embedding for link prediction.
• Constructed a multimodal graph integrating drug-drug, drug-protein, and protein-protein interaction networks.
• Trained and evaluated cell line-specific prediction models.

Main Results:

• The GCN model accurately predicted cell line-specific synergistic drug combinations.
• 30 out of 39 cell line-specific models achieved an Area Under the Curve (AUC) > 0.80, with a mean AUC of 0.84.
• Literature validation confirmed synergistic antitumor activity for many top predicted drug combinations.

Conclusions:

• The GCN model offers a promising computational approach for predicting synergistic drug pairs.
• This study provides a method to optimize drug combinations in silico for cancer therapy.
• The findings pave the way for more efficient and effective cancer treatment strategies.