Research on Network Construction and Application of Novel Drug Repositioning Strategies

---

Background

During the COVID-19 pandemic, researchers and pharmaceutical companies have been dedicated to developing treatments and vaccines. Drug repositioning, due to its shortcut, is considered a rapid and effective response strategy. Drug repositioning attempts to discover new uses for approved drugs and is considered cheaper and quicker compared to traditional drug discovery pathways [1–3]. For example, Remdesivir and Dexamethasone are two successful repositioned drugs [4–6]. Although the global pandemic has gradually moved towards an endemic stage, the virus continues to spread. The importance of quickly discovering candidate drugs and providing them to experts in the medical or pharmaceutical fields for research has been profoundly emphasized by the COVID-19 pandemic [7].

With the advancement of biological mechanisms and the collection of biomedical knowledge, more accurate and precise computational-based drug repositioning has become possible. Network medicine provides candidate drugs by observing the complex relationships between biological entities such as drugs, genes, and diseases [8–11]. However, in the context of newly emerging infectious diseases, pre-accumulated databases may lack sufficient information, making it difficult for network methods to infer new drug repositioning [12, 13].

Source of Research

This paper was written by Yonghyun Nam, Anastasia Lucas, Jae-seung Yun, and other researchers from institutions such as the University of Pennsylvania, Seoul National University, The Catholic University, and Ohio State University. The paper was published in the Journal of Translational Medicine in 2023 [1].

Research Process

Research Steps

1. Construction of Multi-layer Disease-Gene-Drug Network:

   • The background network consists of 591 diseases, 26,681 proteins, and 2,173 drug nodes, and is constructed based on public databases such as the Comparative Toxicogenomics Database (CTD), Search Tool for the Retrieval of Interacting Genes/Proteins (STRING), and DrugBank.
   • By calculating proximity, a disease-disease network, protein-protein interaction network, and drug-drug network are formed.
   • Collect 18 comorbid diseases and 17 related genes associated with COVID-19 to estimate the complementary connections of the COVID-19 node within the underlying network.

2. Network Complementary Method:

   • For new disease nodes, such as COVID-19, a network complementary linkage method was developed to estimate the formation of functional networks.
   • By introducing new disease data and using a graph-based semi-supervised learning algorithm, candidate drugs are predicted and prioritized by score.

3. Drug Screening and Validation:

   • Candidate drugs undergo validation in the electronic health records (EHR) system of the University of Pennsylvania by analyzing drug-patient data.
   • Logistic regression is used to analyze the relationship between candidate drugs and COVID-19-related phenotypes (susceptibility, hospitalization, and severity).

Data Analysis Methods

1. Network Construction:

   • Construct a multi-layer heterogeneous graph, where nodes represent diseases, genes, and drugs, and the intra-layer and inter-layer similarities are represented by coupling matrices.
   • Use the 35 associations collected in the literature for COVID-19 to complete the network.

2. Semi-supervised Learning:

   • Label propagation algorithm where COVID-19 nodes are labeled, generating scores for all other nodes to form a prioritized list of candidate drugs.

3. Prediction Validation:

   • Perform association analysis of orders and infection phenotypes on the patient data registered in the Penn Medicine system.

Main Results

1. Network Completion:

   • After network completion, a total of 1,440,998 associations, including those with COVID-19, were constructed, identifying 35 complementary connections.
   • The completed network reveals the longitudinal evolutionary path during the pandemic and determines the direct connections of related genes and diseases.

2. Drug Prioritization:

   • Network scoring identified 30 candidate drugs, including Dexamethasone and Prednisolone, in the initial score.
   • The graph-based semi-supervised learning algorithm propagates label information based on network structure, assessing the effectiveness of drugs.

3. EHR Validation:

   • EHR data show that 8 candidate drugs are significantly associated with COVID-19 phenotypes, including nonsteroidal anti-inflammatory drugs (NSAIDs) such as Aspirin and Ibuprofen.

Research Conclusions

• Scientific and Application Value:

  • The research demonstrates the potential of network medicine and computational methods in emerging infectious disease drug discovery.
  • The rapid and flexible network completion method effectively links new disease conditions to existing networks, providing candidate drugs for public health emergencies.

• Research Highlights:

  • Innovatively developed network complementary linkage methods to address the issue of new disease nodes lacking connections.
  • The use of a graph-based semi-supervised learning algorithm in drug assessment underscores the value of computational drug repositioning.

• Need for Further Research:

  • Clinical trials are needed to verify the effectiveness and safety of the recommended candidate drugs.
  • Expanding the biomedical network to include more complex relationships (such as single nucleotide polymorphisms and green nanomaterials).

Future research should not only continue exploring drug repositioning methods but also enhance the clinical validation mechanism of its applications to more quickly respond to public health crises.

This research, through the innovative methods of network completion and graph-based semi-supervised learning, demonstrates the ability to rapidly screen and validate candidate drugs, providing important insights and practical value for future responses to emerging infectious diseases.