Deep learning identifies synergistic drug combinations for treating COVID-19

Abstract

Effective treatments for COVID-19 are urgently needed. However, discovering single-agent therapies with activity against severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has been challenging. Combination therapies play an important role in antiviral therapies, due to their improved efficacy and reduced toxicity. Recent approaches have applied deep learning to identify synergistic drug combinations for diseases with vast preexisting datasets, but these are not applicable to new diseases with limited combination data, such as COVID-19. Given that drug synergy often occurs through inhibition of discrete biological targets, here we propose a neural network architecture that jointly learns drug-target interaction and drug-drug synergy. The model consists of two parts: a drug-target interaction module and a target-disease association module. This design enables the model to utilize drug-target interaction data and single-agent antiviral activity data, in addition to available drug-drug combination datasets, which may be small in nature. By incorporating additional biological information, our model performs significantly better in synergy prediction accuracy than previous methods with limited drug combination training data. We empirically validated our model predictions and discovered two drug combinations, remdesivir and reserpine as well as remdesivir and IQ-1S, which display strong antiviral SARS-CoV-2 synergy in vitro. Our approach, which was applied here to address the urgent threat of COVID-19, can be readily extended to other diseases for which a dearth of chemical-chemical combination data exists.

Keywords: SARS-CoV-2; deep learning; drug discovery; drug synergy.

Fig. 1.

ComboNet for synergistic drug combination discovery. (A) ComboNet is composed of two networks: a DTI and a target−disease association network. The antiviral effect of a single drug $p_A$ is predicted from its representation ${\mathit{z}}_A$. The vector ${\mathit{z}}_A$ characterizes the DTI features of drug A. (B) The antiviral effect of a combination is predicted from its representation ${\mathit{z}}_{AB}$, which is computed from the molecular representations of each individual drug ${\mathit{z}}_A,{\mathit{z}}_B$. ComboNet is trained on drug combination synergy, single-drug antiviral activity, and DTI data.

Fig. 2.

In silico evaluation of ComboNet (A) The training, validation, and test set composition for SARS-CoV-2. (B) Results on SARS-CoV-2 drug combination test set. Our full ComboNet model outperforms all other baselines. (C) ROC-AUC plot of ComboNet ensemble on the entire test set. (D) ROC-AUC plot of ComboNet ensemble on the hard drug combinations with at least one new drug. (E) Statistical characteristics of ComboNet ensemble for all the datasets, where “screen” refers to the top 30 candidates we experimentally tested.

Fig. 3.

Discovery of synergistic drug combinations for SARS-CoV-2 (A) Two drug combinations are discovered by our model: remdesivir + reserpine and remdesivir + IQ-1S. (B) Host cell viability matrices show the two drug combinations have low cytotoxicity. (C) Dose–response and Bliss synergy matrices of remdesivir + reserpine. Numbers in the dose–response matrix stand for viral infection rate. Numbers in the Bliss synergy matrix stand for synergy score. Both are the lower the better. (D) Dose–response and Bliss synergy matrices of remdesivir + IQ-1S. (E) The correlation between predicted ranking and DBSumNeg score (lower DBSumNeg means more synergistic). (F) The t-distributed stochastic neighbor embedding visualization (36) of the chemical space explored across the training set, test set, and experimentally validated combinations.