[Paper Review] Predicting potential drug targets and repurposable drugs for COVID-19 via a deep generative model for graphs

Sumanta Ray, Snehalika Lall+3 more|arXiv (Cornell University)|Jul 5, 2020

Computational Drug Discovery Methods59 references11 citations

TL;DR

This study proposes a deep generative model using Variational Graph Autoencoders (VGAE) to predict novel drug–host protein interactions for SARS-CoV-2, integrating three curated biological networks: drug–protein, human protein–protein, and SARS-CoV-2–human protein interactions. The model identifies high-probability drug–host links, revealing repurposable drugs for host-directed therapy with strong predictive accuracy and computational efficiency.

ABSTRACT

Coronavirus Disease 2019 (COVID-19) has been creating a worldwide pandemic situation. Repurposing drugs, already shown to be free of harmful side effects, for the treatment of COVID-19 patients is an important option in launching novel therapeutic strategies. Therefore, reliable molecule interaction data are a crucial basis, where drug-/protein-protein interaction networks establish invaluable, year-long carefully curated data resources. However, these resources have not yet been systematically exploited using high-performance artificial intelligence approaches. Here, we combine three networks, two of which are year-long curated, and one of which, on SARS-CoV-2-human host-virus protein interactions, was published only most recently (30th of April 2020), raising a novel network that puts drugs, human and virus proteins into mutual context. We apply Variational Graph AutoEncoders (VGAEs), representing most advanced deep learning based methodology for the analysis of data that are subject to network constraints. Reliable simulations confirm that we operate at utmost accuracy in terms of predicting missing links. We then predict hitherto unknown links between drugs and human proteins against which virus proteins preferably bind. The corresponding therapeutic agents present splendid starting points for exploring novel host-directed therapy (HDT) options.

Motivation & Objective

• To address the urgent need for rapid therapeutic strategies against SARS-CoV-2 by identifying repurposable drugs with low side-effect risk.
• To overcome limitations in direct antiviral drug development due to unknown 3D structures of viral proteins and potential resistance.
• To leverage large-scale, curated biological interaction networks to improve prediction of drug–host protein interactions.
• To develop a scalable, deep generative model that predicts missing links in complex host–virus–drug interaction networks.
• To identify novel host-directed therapy (HDT) targets by predicting interactions between drugs and human proteins that SARS-CoV-2 exploits.

Proposed method

• Constructs a heterogeneous network integrating three data sources: drug–target interactions, human protein–protein interactions, and SARS-CoV-2–human host protein interactions.
• Applies a scalable Fast Variational Graph Autoencoder (FastVGAE) to learn low-dimensional latent representations of nodes in the graph using graph convolutional networks (GCNs).
• Employs a node sampling strategy with degree-based probability to reconstruct a subgraph at each training iteration, reducing computational cost.
• Uses a stochastic gradient descent-based optimization to maximize the likelihood of reconstructing the adjacency matrix from latent node representations.
• Applies a sigmoid function to decode reconstructed adjacency matrices, estimating link probabilities between drug and host protein nodes.
• Utilizes biased random walks (via Node2vec) with parameters p and q to generate feature representations for nodes, enabling effective neighborhood encoding.

Experimental results

Research questions

• RQ1 Which drugs are most likely to interact with human host proteins that SARS-CoV-2 targets, suggesting potential host-directed therapy candidates?
• RQ2 How accurately can a deep generative model predict missing links in a heterogeneous host–virus–drug interaction network?
• RQ3 Can integrating long-standing curated biological networks with newly published SARS-CoV-2–host protein interaction data improve drug repurposing predictions?
• RQ4 What is the performance of a FastVGAE model in predicting novel drug–host protein interactions compared to traditional network-based approaches?
• RQ5 Which specific drugs emerge as top candidates for repurposing based on predicted high-probability interactions with host proteins critical for viral replication?

Key findings

• The model achieves high accuracy in predicting missing links in the interaction network, validated through reliable simulations.
• The FastVGAE framework significantly reduces computational time compared to standard VGAE, enabling scalable analysis of large biological networks with 16,444 nodes.
• The method successfully identifies novel, high-probability links between drugs and human host proteins that SARS-CoV-2 interacts with, suggesting new host-directed therapy candidates.
• The top-ranked predicted drug–host protein interactions provide strong starting points for experimental validation of repurposed drugs.
• The integration of three distinct, long-curated biological networks—drug–protein, human PPI, and SARS-CoV-2–host PPI—yields a comprehensive and biologically meaningful network for prediction.
• The use of degree-based node sampling with α=2 enhances model stability and convergence by prioritizing high-degree nodes during subgraph reconstruction.