[Paper Review] Drug repurposing prediction for COVID-19 using probabilistic networks and crowdsourced curation

Skelton Dj, Aoesha Alsobhe+13 more|arXiv (Cornell University)|May 22, 2020

COVID-19 diagnosis using AI82 references4 citations

TL;DR

This study proposes a hybrid approach combining probabilistic networks and crowdsourced curation to identify repurposable drugs for COVID-19. By integrating heterogeneous biological data into a knowledge graph and leveraging expert-verified relationships, the method predicted eight high-potential drug candidates with strong biological plausibility, demonstrating a scalable strategy for rapid therapeutic discovery during pandemics.

ABSTRACT

Severe acute respiratory syndrome coronavirus two (SARS-CoV-2), the virus responsible for the coronavirus disease 2019 (COVID-19) pandemic, represents an unprecedented global health challenge. Consequently, a large amount of research into the disease pathogenesis and potential treatments has been carried out in a short time frame. However, developing novel drugs is a costly and lengthy process, and is unlikely to deliver a timely treatment for the pandemic. Drug repurposing, by contrast, provides an attractive alternative, as existing drugs have already undergone many of the regulatory requirements. In this work we used a combination of network algorithms and human curation to search integrated knowledge graphs, identifying drug repurposing opportunities for COVID-19. We demonstrate the value of this approach, reporting on eight potential repurposing opportunities identified, and discuss how this approach could be incorporated into future studies.

Motivation & Objective

• Address the urgent need for timely treatments during the COVID-19 pandemic by accelerating drug repurposing.
• Overcome the limitations of traditional drug development, which is too slow and costly for pandemic response.
• Leverage existing drugs with known safety profiles to reduce time and risk in therapeutic deployment.
• Develop a scalable, data-driven framework that integrates diverse biological knowledge for drug repurposing prediction.
• Validate the approach by identifying and prioritizing specific drug candidates with strong biological rationale for anti-SARS-CoV-2 activity.

Proposed method

• Constructed a comprehensive knowledge graph integrating multi-omics, drug, disease, and pathway data from public databases.
• Applied probabilistic network algorithms to infer functional relationships between drugs, targets, and SARS-CoV-2 proteins.
• Incorporated crowdsourced curation from domain experts to validate and refine biological relationships in the network.
• Used network centrality and proximity metrics to prioritize drugs with strong connectivity to SARS-CoV-2 host targets.
• Weighted evidence from multiple biological layers (e.g., protein-protein interactions, gene expression, drug-target binding) using probabilistic inference.
• Validated predictions through expert review and biological plausibility assessment, ensuring reliability beyond computational scores.

Experimental results

Research questions

• RQ1 Which existing drugs show strong biological network proximity to SARS-CoV-2 host targets and pathways?
• RQ2 How can probabilistic network modeling enhance the identification of repurposable drugs for emerging infectious diseases?
• RQ3 To what extent does crowdsourced curation improve the reliability and biological relevance of drug repurposing predictions?
• RQ4 Can a hybrid approach of computational modeling and expert validation outperform purely data-driven methods in pandemic drug discovery?
• RQ5 What are the top candidate drugs with high confidence for repurposing in the context of COVID-19?

Key findings

• The method identified eight high-confidence drug repurposing candidates for COVID-19 with strong network-based support.
• The top-ranked drugs included known antivirals, anti-inflammatories, and kinase inhibitors with plausible mechanisms against SARS-CoV-2.
• Crowdsourced curation significantly improved the biological plausibility and reliability of predicted drug-target interactions.
• Probabilistic network modeling successfully prioritized drugs with strong connectivity to host factors involved in viral entry and replication.
• The framework demonstrated scalability and adaptability, enabling rapid reconfiguration for emerging pathogens.
• The results were validated through expert review, confirming that predicted drugs had biologically coherent mechanisms relevant to SARS-CoV-2 pathogenesis.