Deep learning identifies synergistic drug combinations for treating COVID-19

Area of Science:

• Computational biology
• Drug discovery
• Virology

Background:

• Developing effective treatments for COVID-19 (caused by SARS-CoV-2) is critical.
• Single-agent therapies have faced challenges, highlighting the need for combination therapies.
• Existing deep learning methods for drug synergy prediction struggle with new diseases like COVID-19 due to limited combination data.

Purpose of the Study:

• To develop a novel neural network architecture for predicting synergistic drug combinations.
• To address the limitations of existing methods in diseases with sparse combination data.
• To identify effective drug combinations against SARS-CoV-2.

Main Methods:

• Proposed a neural network with two modules: drug-target interaction and target-disease association.
• Integrated drug-target interaction data, single-agent antiviral activity, and limited drug-drug combination data.
• Leveraged additional biological information to improve synergy prediction accuracy.

Main Results:

• The proposed model significantly outperformed previous methods in synergy prediction accuracy with limited data.
• Empirically validated predictions, identifying two synergistic drug combinations against SARS-CoV-2 in vitro.
• Discovered remdesivir and reserpine, and remdesivir and IQ-1S as potent antiviral combinations.

Conclusions:

• The novel neural network approach effectively predicts drug-drug synergy, even with limited combination data.
• Identified promising synergistic drug combinations for potential COVID-19 treatment.
• The methodology is adaptable for discovering combination therapies for other diseases with scarce data.