COVID-19 Drug Repurposing: A Network-Based Framework for Exploring Biomedical Literature and Clinical Trials for Possible Treatments

Ahmed Abdeen Hamed, Tamer E. Fandy, Karolina L. Tkaczuk, Karin Verspoor, Byung Suk LeePharmaceutics 2022 · DOI

TL;DR

This paper introduces a novel network-based computational framework to identify potential COVID-19 drug combinations by analyzing biomedical literature and clinical trial data. By looking for overlapping "cliques" of associated drugs in both sources, the authors successfully pinpointed several promising multi-drug treatment candidates, validating their hypothesis that existing FDA-approved drugs can be repurposed for COVID-19.

What Problem Does This Paper Address?

The COVID-19 pandemic created an urgent need for effective treatments. While vaccine development was accelerated, finding and validating new therapeutic options for infected individuals remained a critical challenge. Drug repurposing, the process of identifying new uses for existing drugs, offered a faster path to potential treatments. However, identifying effective combinations of drugs, which can often be more potent than single agents, is complex. This paper addresses the challenge of systematically discovering and validating such drug combinations by leveraging the vast and growing body of biomedical literature and clinical trial data. The authors hypothesize that a drug combination is a strong candidate for COVID-19 treatment if it is supported by evidence in both the scientific literature (co-occurrence of drugs in publications) and clinical trials (drugs being tested together).

Methods

The core of this research is a computational framework designed to identify drug combinations. It involves several key steps:

  • Drug Name Extraction: Using the ChEBI ontology, the framework extracts drug names from the abstracts of biomedical publications and the intervention/medication sections of clinical trial records.
  • Association Analysis and Network Construction: The extracted drug names are used to build association networks. The Apriori algorithm is employed to identify significant co-occurrences of drugs within publications, filtering out weak associations and creating a network where drugs are nodes and associations are links. This process is repeated for clinical trial data.
  • Clique Detection: In graph theory, a clique is a subset of nodes where every pair of nodes is connected by an edge. The framework applies a maximal clique algorithm to identify these densely connected groups of drugs within both the literature-derived and clinical trial-derived networks. The idea is that drugs within a clique are strongly associated with each other.
  • Validation and Discovery (Search-n-Match Algorithm): This is the novel validation step. The algorithm compares cliques of the same size found in the literature network with those found in the clinical trial network. If a clique from the literature has a corresponding clique of the same size in the clinical trials, it suggests a potential treatment candidate. The similarity is quantified using a modified Jaccard index. The intersection of these matching cliques represents the identified drug combinations.

Datasets

The study utilized two primary datasets:

  • Biomedical Publications: A collection of approximately 110,000 publications related to "COVID-19" from PubMed, with a cutoff date of June 25, 2021.
  • Clinical Trial Records: A set of approximately 10,000 COVID-19 clinical trial records, also extracted using the search query "COVID-19" and with the same cutoff date. These were sourced from ClinicalTrials.gov.

The paper does not introduce a new benchmark dataset but rather leverages existing, publicly available data sources.

Key Insights and Ideas

  • The Power of Overlap: The central idea is that a drug combination is more likely to be effective if it appears as a strongly associated group in both scientific literature and clinical trials. This "overlap" provides a robust signal for potential therapeutic value.
  • Cliques as Communities: Identifying cliques in the drug association networks is a clever way to represent groups of drugs that are frequently discussed or tested together. This network structure helps to uncover complex relationships beyond simple pairwise associations.
  • Clinical Trials as Ground Truth: Using clinical trial data as a validation mechanism is crucial. Clinical trials represent a higher bar of evidence, as they involve actual testing of drug interventions. Matching literature-based findings to clinical trial data significantly increases confidence in the identified candidates.
  • Beyond Single Drugs: The research emphasizes that effective COVID-19 treatments might involve combinations of drugs, rather than single agents. This is particularly relevant given the virus's ability to mutate and the potential for synergistic effects between different drugs.

Key Results

The framework successfully identified potential COVID-19 treatment candidates in the form of two, three, and four-drug combinations.

  • Two-Drug Combinations: Table 5 lists several pairs of drugs that were identified as potentially combinable, with discussions on their "combineability" based on biochemical and pharmacological properties. Some of these, like lopinavir and ritonavir, are already FDA-approved combinations.
  • Three and Four-Drug Combinations: The study identified several cliques of size three and four that had matching components in both literature and clinical trials. For example, the combination of hydroxychloroquine, lopinavir, and favipiravir was identified as a potential three-drug treatment.
  • Validation Statistics: Table 7 provides statistics on clinical trials investigating various drug combinations, showing the number of trials and their status (recruiting, completed, etc.). This table supports the findings by demonstrating that many of the identified combinations are indeed under active investigation.
  • Significant Reduction in Candidates: The comparison between literature-derived cliques and clinical trial-derived cliques significantly reduced the number of potential candidates, highlighting the importance of the validation step. For instance, out of 3643 cliques detected from publications, only a subset was validated against clinical trials.

Limitations and Future Work

  • Literature Coverage: The authors acknowledge that their analysis was based on a specific strategy for collecting COVID-19 literature. They suggest that using more comprehensive datasets (e.g., CORD-19, LitCovid) could yield broader insights.
  • Therapy for Severe vs. Non-Severe Cases: The current analysis did not differentiate between potential treatments for severe versus non-severe COVID-19 cases.
  • Predicting Potential Therapy: The framework focuses on identifying drug combinations based on existing evidence and does not predict novel therapeutic mechanisms or outcomes.
  • Future Work: The authors propose identifying additional validation sources for cliques that lacked coverage in clinical trial records. They also plan to investigate how these drug combinations might affect patients with pre-existing conditions like asthma, diabetes, or hypertension. Further focus will be placed on understanding the "combineability" of identified drug pairs and triples.

Summary updated

Abstract

Our framework has identified treatments in the form of two, three, or four drug combinations (e.g., hydroxychloroquine, doxycycline, and azithromycin). The identifications of the various treatment candidates provided sufficient evidence that supports the trustworthiness of our hypothesis.
BibTeX
@inproceedings{hamed2022covid,
  author = {Ahmed Abdeen Hamed and Tamer E. Fandy and Karolina L. Tkaczuk and Karin Verspoor and Byung Suk Lee},
  title = {{COVID-19 Drug Repurposing: A Network-Based Framework for Exploring Biomedical Literature and Clinical Trials for Possible Treatments}},
  booktitle = {Pharmaceutics},
  year = {2022},
  doi = {10.3390/pharmaceutics14030567},
}
Figure