Bioinformatics tools for drug repurposing: a tutorial using heart failure as a case study

Abstract

Purpose

Drug repurposing is a crucial strategy for researchers worldwide to accelerate drug development and mitigate associated risks and costs. Heart failure (HF) is a major global health problem with high prevalence and mortality rates. There are significant sex differences in HF, including in the risk factors and phenotype, which demand a sex-personalized drug treatment. A convenient approach in that direction is the reuse of drugs already approved for other conditions that are known to interact in sex-biased dysregulated pathways in HF. Numerous bioinformatics tools can help identify those candidates. This tutorial explores the utility of specific bioinformatics tools in identifying drugs to treat HF as a case study.

Methods

Herein, we explain how NeDRex helps identify genes associated with disease and drug repurposing and how functional enrichment analysis can be performed with ShinyGO. We also explain how to predict targets of small bioactive molecules with SwissTargetPrediction and how to retrieve known and predicted interactions between chemicals and proteins with STITCH.

Results

The tutorial demonstrates the use of these tools in searching for new HF treatments.

Conclusion

This tutorial is designed to ease entry into the utilization of the mentioned bioinformatics tools. This approach can also set a precedent for applying such tools to other diseases. The results presented in this tutorial are illustrative and do not constitute definitive evidence. They are intended for demonstration purposes only.

Graphical abstract

Utilization of bioinformatics tools to uncover novel insights, facilitating the development of personalized medicine strategies for the treatment of heart failure.

1. Introduction

The traditional drug design and development process is time-consuming and very expensive. In contrast, drug repurposing is a hopeful substitute because it recognizes new applications for drugs approved by regulatory agencies such as the FDA and EMA [1]. The pressing urgency in responding to the COVID-19 pandemic increased awareness about the importance of drug repurposing [2,3]. The basis of drug repurposing rests upon the premise that similar molecular mechanisms may be involved in different diseases [4]. Today, it is becoming imperative to reduce both the costs associated with drug discovery and development and the time elapsed from the laboratory research to the patient administration, producing at the same time target-specific drugs with minimal side effects [5,6].

Historically, fertile women have not been included in clinical trials to protect their well-being and offspring. Consequently, male physiology became the reference in clinical research and drug development. However, women experience more potent side effects, greater adverse reactions, and toxicity with specific treatments. In response to this concern, in 2014, NIH mandated researchers to incorporate sex as a biological variable in research design [7]. Therefore, sex should be taken into account in repurposing so that drugs are similarly safe and effective for women and men [8]. Bioinformatics can contribute significantly to the selection of drug candidates that are more efficacious in one sex or to discarding drug candidates predicted to cause unintentional therapeutic effects that are more probable to occur in one sex, accelerating the research process and reducing its costs [5,6,9].

Understanding the binding mode between a small bioactive molecule such as a drug and proteins is essential to comprehending the molecular mechanisms related to the drug effects. It can lead to identifying new disease therapeutic targets [10,11]. For example, based on the idea that similar ligands are likely to interact with similar proteins, bioinformatics tools can predict the protein targets of a bioactive small molecule [11]. Numerous commercial software packages are available; however, we will focus on those freely available online and most widely utilized in health sciences. We prioritized user-friendly and intuitive interface tools to engage the most diverse audience possible. We will use Cytoscape [12], a powerful software platform widely applied by researchers for creating and visualizing complex networks and integrating these with any attribute data. The possibility of running in different operating systems, including Windows, macOS, and Linux, significantly contributes to its accessibility to a wide range of users. Cytoscape allows the handling of large-scale networks, providing high-quality visualizations of those and presenting customization options for node and edge styles to enhance the interpretability of networks. Most interestingly, Cytoscape has several plugins (apps) that extend its functionality, allowing users to add features for specific analyses, visualizations, or data formats. In addition to visualization, this tool includes various analytical tools for network analysis, such as clustering, pathway analysis, and network statistics, which can help extract meaningful insights from data. Considering all these features, we preferred tools that automatically integrate into Cytoscape. Specifically, using the NeDRexApp (Network-based Drug Repositioning and Drug Response Exploration) [13], a very recent application from Cytoscape, drug candidates for repurposing in heart failure (HF) will be identified, and the rationale behind their use will be discussed. Additionally, we will show how to extract the biological processes and pathways associated with our gene/protein targets using the tool ShinyGO [14]. Also, the bioinformatics tools SwissTargetPrediction [15] and STITCH [16] will provide information to better understand a given drug's molecular mechanisms of action. For this, we can use an experimental dataset, literature search, or databases such as Proteomics Identifications (PRIDE) [17] or Gene Expression Omnibus (GEO) [18]. Although this tutorial shows a joint approach to utilizing different tools, they can be used independently (Fig. 1). These computational approaches can help identify specific drug candidates that can be prioritized for experimental applications.

Fig. 1.

Overall workflow of the tutorial.

HF will be used as a case study for this tutorial to illustrate this bioinformatics approach. HF is considered a global pandemic, characterized by significant morbidity and mortality, poor functional capacity and quality of life, and high healthcare costs. The general longer life expectancy of the population and improved survival after diagnosis, related to the availability of evidence-based treatments, have been associated with the anticipated increase in HF prevalence [19,20]. Sex-based differences have been reported in HF prevalence, risk factors, and clinical presentation [21,22]. For instance, men are predisposed to HF with reduced ejection fraction (HFrEF), while women prevail among patients with HF with preserved ejection fraction (HFpEF) [23]. Despite commendable efforts made in the treatment of HF, it is daunting to witness the persistence of complications experienced by many patients, hindering their recovery and well-being. This sobering reality underscores the urgent need to delve deeper into the intricate molecular targets of drug action, with a particular focus on understanding the profound impact of sex differences. By doing so, we can pave the way for the development of sex-targeted therapy, leading to improvements in patient outcomes. The reader should be aware that the analysis presented in this tutorial constitutes a mere exercise that was not based on an extensive literature search. Therefore, none of the outputs should be viewed as preliminary evidence for drug repurposing in HF.

2. Utilization of the NeDRex app to explore drugs for heart failure treatment

NeDRexApp is a Cytoscape application integrating disease, gene, protein, and drug associations. This app can be helpful in the identification of disease modules, i.e., sub-networks of genes that participate in relevant pathways in a given disease, thus containing potential drug targets [13]. NeDRex is the first generically applicable integrated platform for network-based disease module discovery and drug repurposing (https://apps.cytoscape.org/apps/nedrex). To learn more about using this app, the reader can consult the NeDRex website (https://nedrex.net/), where some tutorial videos can be found (https://nedrex.net/media.html). Next, the workflow for NeDRexApp utilization in the Cytoscape platform is explained to demonstrate its usefulness in summarizing, through a network, the drugs typically prescribed for a given condition, in this case, HF. The reader is referred to the “Supplementary tutorial slides”, which contain supporting screenshots to follow the tutorial more easily.

• 1.In Cytoscape, click on Apps and next on App Manager. In the search line, search for NeDRex and proceed to the app installation.

  NOTE: Cytoscape version 3.9.1 and NeDRexApp version 2.0.0 were used in this tutorial.

• 2.Click on "File", go to "Import", and click on "Network from Public Databases". On "Data Source", select "NeDRex: network query from NeDRexDB". Next, choose the associations you need for your analysis in "Association Options". For our goal, select Gene-Disorder, Gene-Protein, Protein-Protein, Drug-Protein, and Disorder-Disorder (associations representing the disorder hierarchy in Mondo Disease Ontology (https://www.ebi.ac.uk/ols/ontologies/mondo)). To learn more about each association type, consult https://nedrex.net/tutorial/availableFunctions.html. In "Protein-Protein Options", do not select any option. Within "Gene-Disorder Options", select OMIM associations and DisGeNET associations. In this last option, set the score value to 0.5. This score refers to the DisGeNET score for gene-disease associations (GDAs) measuring the level of evidence supporting the GDA (the number of sources that report the association, the type of curation for each of the sources, the animal models where the association has been studied, and the number of supporting publications from text-mining based sources). The score ranges from 0 to 1. A higher score is associated with stronger evidence supporting the association [24]. We will select an intermediate DisGeNET cutoff score of 0.5 for this tutorial. Select Approved, Experimental, Investigational, Vet_approved, and Nutraceutical in "Drug Options". In "Taxonomy", select Human. In the "Network name" box, give a designation to the network, for example, Network_1. Check “I agree with the NeDRex Terms of Use” in the "License Agreement". Next, click on "Import". Your network will be downloaded and created in Cytoscape. This process can take a few minutes. When asked if you want to build a view of the network, click “Cancel”. This way, the Network will be imported, but the Network View will not be depicted, saving computer memory.
• 3.Next, go to https://www.ebi.ac.uk/ols/ontologies/mondo to find the term that best identifies the condition at scope. In this case, for HF, you will find the term MONDO:0005252.
• 4.After downloading the NeDRex database, keep the imported network selected, click on "Apps", select "NeDRex" in the new window, and click on “Quick Select”. In the new box, “Choose the node type you want to select from:”; select in this case, Disorder. Then, write in “Type in the disease/drug/gene/protein/pathway name or ID:” the term mondo.0005252, which corresponds to HF. Select the corresponding term and click on “Add to the list”. The term will appear in the box “Added”. Then, click on “Select in Network” and close the box.
• 5.Click on "Apps", then go to "NeDRex", and click on “Get Disease Genes” to obtain the genes associated with the selected disorder, in this case, HF. The genes are obtained from the databases integrated into NeDRexDB. Since heart failure includes other conditions in MONDO, check the box “Include all subtypes of disorders”. To define a name for the resulting network, check on “Use custom name for the result network” and write HF_subtypes_associated genes in the field “Name of the result network”. Then, click "OK" to obtain the network from Fig. 2A. According to NeDRex visual style, the square blue nodes represent genes, and the diamond pink ones correspond to the disorders. You can obtain this style by clicking on "Apps", selecting "NeDRex", and clicking on “Create NeDRex visual style”.
• 6.To save the network image, click on the icon to Detach View (white arrow in a black square) in the Cytoscape Network View Window and then maximize the window. Click on the icon to Fit Content (a magnifying glass with two arrows). This last option will allow you to zoom out of the current view and display all its elements. Next, go to "File" > "Export" and click on “Network to Image”. Select PNG in the “Export File Format” and then choose the name of the image and the preferred directory. If you wish to obtain publication-quality figures, in the Image Size, select the maximum percentage (500 %); in Units, select inches; and in Resolution (DPI), select the maximum (600). Then, click on the icon to Reattach View (a pin). However, you can save the image in formats like PDF and SVG.

  NOTE: Remember always to record the date of your analysis since results might change due to application and database updates.

• 7.Next, to obtain drugs related to the HF-associated genes, go to "Apps" > "NeDRex" > "Select Nodes" and click on “All nodes of specific type”. In "Node type", select Gene and click "OK". All genes in the network from Fig. 2A will appear selected (in yellow).
• 8.Go to "Apps" > "NeDRex" > "Drug Prioritization" and click on “All drugs targeting the selection”. This last option was selected because we do not intend to rank the drugs but obtain all drugs targeting the selection. Then, select “Include only approved drugs”, and write HF_subtypes_associated genes_All drugs approved as the name of the resulting network after clicking on “Use custom name for result network”. Click "OK" (Fig. 2B). In this network, hexagon nodes represent drugs.
• 9.To change the color of the drugs according to the drug group, click on the Style icon (paintbrush) on the control panel's left. On "Node" properties, in "Fill Color", click on the second square. In the "Column", choose drugGroups and then select, for instance, green (by clicking twice in front of the group of drugs and clicking in the three-point rectangle) for approved drugs, orange for approved, investigational and approved, experimental, and gray for the other options. Click on "OK".

  NOTE: Drugs are categorized by group, which determines their drug development status (according to DrugBank). Consult https://dev.drugbank.com/guides/terms/drug-group to learn more about the meaning of each drug group.

• 10.To save the image of the network, proceed as described in point 6.
• 11.To learn more about the diseases, genes, proteins, and drugs in the networks, you can right-click on the nodes, go to "NeDRex", and click on “Open Entry in Database”. A box will appear with the links to databases. Click on the link to learn about that drug, gene, or condition.

Fig. 2.

Network of the genes associated with heart failure and other disorder subtypes, according to MONDO and NeDRex databases (A). Network of the drugs associated with the genes related to heart failure (B). Blue square nodes represent genes, pink diamond ones correspond to the disorders, and hexagon nodes correspond to drugs (green-approved; orange-approved, investigational and approved, experimental; gray-other group drugs). Databases sourced on January 9th, 2025. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

3. Step-by-step analysis of the pathways related to the genes associated with heart failure using ShinyGO

Identifying the molecular pathways and functional categories under gene ontology (GO) is essential after retrieving the genes associated with a given condition, in this case, HF. This can be particularly relevant to assign a putative role for such genes within pathways and processes related to the condition at scope. The online tool ShinyGO v0.81 (http://bioinformatics.sdstate.edu/go/) was used for this purpose. ShinyGO is a web-based tool designed for gene ontology enrichment analysis and visualization. This tool was selected due to a combination of several features. These include the possibility of accessing a wide range of databases, such as KEGG [25], Reactome [26,27], PANTHER [28,29], GO Biological Process, GO Molecular Function, GO Cellular Component [30,31], and STRING [32], among others related to diseases and drugs. ShinyGO combines multiple analyses, including GO enrichment, pathway, and protein-protein interaction analyses. This integrated approach allows researchers to gain more comprehensive insights from their data without switching between different tools. Another helpful feature is the possibility of customizing analysis parameters to fit specific needs and adjusting the graphic presentation of the results. Also, this tool allows real-time visualization of results, permitting users to explore and understand their data dynamically. This includes interactive plots and charts that make it easier to interpret complex data. ShinyGO can generate results in multiple formats, easily shared or incorporated into publications. Other tools that can give us a similar output include ClueGO [33], which, despite having a direct plugin in Cytoscape, can result in very dense networks, especially if the input is very large or redundant; enrichGO, which requires knowledge of basic programming in R; and g:Profiler [34], whose analysis output is limited visually, being primarily table-based. Although ShinyGO does not have a direct plugin in Cytoscape, it allows users to download the nodes and edges and import them as a network into Cytoscape.

The workflow for the analysis in ShinyGO is as follows:

• 1.Go to the ShinyGO website (http://bioinformatics.sdstate.edu/go/).
• 2.We will work with the species Human for this tutorial. Keep it selected.
• 3.Paste your list in the box indicated for genes. In this case, we will paste the genes obtained in the analysis depicted in Fig. 2A.
• 4.Keep the Background gene list. By default, ShinyGO compares our gene list with a background of all protein-coding genes in the genome. Nevertheless, using a tailored background gene list that incorporates protein-coding genes from a particular tissue sample, such as cardiac tissue, may yield more biologically relevant enrichment outcomes.
• 5.For this analysis, we will maintain the default settings. Keep the “FDR cutoff” as 0.05, the “number of pathways to show” as 20, the “Pathway size. Min.” as 2, and the “Max.” as 5000. The selection of the option “Abbreviated pathways” will shorten the name of the depicted pathways. The check on “Remove redundancy” will eliminate similar pathways that share 95 % of their genes and 50 % of the words in their names, keeping only the pathway with the highest significance. We suggest you keep the default settings in the first analysis, and after submitting, in the tab “Enrichment”, you can see important information about the methods and the interpretation of the results that can help you to change some settings if you need to. Click on "Submit".
• 6.On the right, in the tab “Chart”, a graphical representation of the analysis can be found. By default, a graphic with the KEGG pathways is shown. Still, on the left, in the item “Pathway database”, other options can be selected, such as GO Biological Process, GO Cellular Component, and GO Molecular Function. You can change the parameters below the graph as desired. The chart can be downloaded by clicking on “Download Plot” (Fig. 3). Other tabs are presented. For instance, in the tab “Enrichment”, a table with the enrichment analysis is shown, and below, relevant information regarding the methods and interpretation of the results. To save the pathways shown in the table, click on “Top pathways shown above”, and a file in CSV format will be downloaded. If you want to save all the pathways, click on “Results on All Pathways”. In the tab “Tree”, a hierarchical clustering tree summarizing the correlation between significant pathways listed in the Enrichment tab is obtained. In the tab “Network”, an interactive network is shown, also representing the relationship between enriched pathways. Within the tab “KEGG”, in the option “Select a significant KEGG pathway to show a diagram with your genes highlighted in red”, it is possible to select a pathway and see the pathway diagram from KEGG below, as well as download it. In the tab “Genes”, a table with information about the genes is shown. Furthermore, in the tab “STRING”, a protein-protein network is retrieved based on the data from STRING-db.

Fig. 3.

Chart of the enriched KEGG pathways associated with the HF-related genes (genes from Fig. 2A). The analysis was made with ShinyGO 0.80 software (http://bioinformatics.sdstate.edu/go/). Databases sourced on January 13th, 2025.

4. Utilization of the NeDRex app to investigate potential drug candidates for repurposing in heart failure

Next, to obtain drugs with potential for repurposing, we will use the DIAMOnD algorithm [35]. This algorithm allows us to expand our network to get not only drugs that can act on our initially selected genes (known as disease genes (or seeds) - directly associated with the disorder at scope) but also drugs that can act on genes associated with those initial seeds and that are indirectly associated with our disorder. To know more about the DIAMOnD algorithm, consult https://nedrex.net/tutorial/methods.html.

• 1.With the network HF_subtypes_associated genes selected, go to "Apps" > "NeDRex" > Select "Nodes" and click on “All nodes of specific type”. In "Node type", select Gene and click "OK". All genes in the network will appear selected (in yellow).
• 2.Go to "Apps" > "NeDRex" > "Disease Module Identification" and click on "Run DIAMOnD". In the number of iterations, maintain 200; in the "Weight of seeds", keep 1 as the default setting. The number of iterations determines the size of the resulting subnetwork. This parameter establishes the intended number of DIAMOnD genes/proteins or number of iterations (genes). It is advisable to choose between 20 and 200. The parameter “Weight of seeds” establishes the importance or impact of initial seed nodes during the algorithm's processing. The DIAMOnD algorithm will result in a new network that includes the initial seeds and the genes identified by DIAMOnD. Keep the option “Return all edges in the result disease module” unchecked. If checked, all edges among genes in the resulting disease module will be included, while unchecking it will only return the edges between seeds and the newly identified nodes. The latter is especially recommended to improve readability whenever disease modules are large. Check on “Use custom name for the result network” and write HF_subtypes_DIAMOnD in the box “Name of the result network” and click "OK". A new network will emerge (Fig. 4), with the previously selected nodes (seeds) in yellow.
• 3.You can change the color of the primary nodes to differentiate from the added ones. To do this, maintain the primary nodes selected and click on the "Style" icon. On "Node" properties, in "Fill Color", right-click on the third square, then "Set Bypass" and choose a color. Click "OK".
• 4.To save the image of the network, proceed as described in point 6 from Section 2.
• 5.Next, with the network from Fig. 4 selected in the control panel of Cytoscape, go to "Apps" > "NeDRex" > "Select Nodes" and click on “All nodes of specific type”. In "Node type", select Gene and click "OK". All genes in the network will appear selected (in yellow).
• 6.Go to "Apps" > "NeDRex" > "Drug Prioritization" and click on “Rank drugs with Closeness Centrality”. Select “Include only approved drugs”, set "Result size" to 50, and rename the resulting network as HF_subtypes_DIAMOnD_Closeness Centrality (Fig. 5). The selected primary seeds will be shown in yellow. Maintain the selection and bypass the color to, for example, yellow. Closeness centrality is a node centrality measure that prioritizes the nodes in a network based on the lengths of their shortest paths to all other nodes in the network. Drugs that are at a close distance to the selected nodes in the disease module have a higher probability of being good candidates for repurposing. A score is presented in the Node Table, and the higher the value, the higher the probability of being good candidates for repurposing. To learn more about this measure, check https://nedrex.net/tutorial/methods.html.
• 7.To save the image of the network, proceed as described in point 6 from Section 2.

Fig. 4.

Network of the disease module for heart failure obtained using the DIAMOnD algorithm. Yellow nodes represent the genes previously selected, and the blue nodes represent the added ones. Databases sourced on January 9th, 2025. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 5.

Network of the drugs associated with the genes related to heart failure. Yellow square nodes represent the genes primarily selected, and the blue nodes are the added ones after the DIAMOnD algorithm application. Hexagon nodes correspond to drugs (green-approved; orange-approved, investigational and approved, experimental; gray-other group drugs). The dark blue node corresponds to niclosamide. Databases sourced on January 9th, 2025. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

NeDRex can also be used to obtain new purposes for drugs indicated for diseases known to be associated with the disease of interest, in this case, HF. We intend to show that, using this approach, we can find drugs with recommendations for diabetes mellitus and obesity as new potential treatments for HF and/or find drugs used in HF that can be relevant to diabetes mellitus and obesity.

• 8.Select the Network_1 in the control panel of Cytoscape, go to "Apps" > "NeDRex", and click on “Quick Select”. In the “Choose the node type you want to select from:”, select Disorder from the menu “<Select type>”. In the box below “Type in the disease/drug/gene/protein/pathway name or ID:”, write mondo.0005015, corresponding to diabetes mellitus. Select the corresponding term and click on “Add to the list”. The term will appear in the box “Added”. Then, click on “Select in Network” and close that box.
• 9.Click on "Apps", then select "NeDRex", and click on “Get Disease Genes” to obtain the genes associated with the selected disorder, in this case, diabetes mellitus (based on databases integrated into NeDRexDB). Since “diabetes mellitus” includes other conditions in MONDO, check the box “Include all subtypes of disorders”. Rename the resulting network as DM_subtypes_associated genes, as explained before. Then, click "OK" to obtain Fig. 6.
• 10.To save the image of the network, proceed as described in point 6 from Section 2.
• 11.Next, to obtain the disease module, go to "Apps" > "NeDRex" > "Select Nodes" and click on “All nodes of specific type”. In "Node type", select Gene and click "OK". All genes in the network of Fig. 6 will be selected (in yellow).
• 12.Go to "Apps" > "NeDRex" > "Disease Module Identification" and click on "Run DIAMOnD" using default settings. Check on “Use custom name for the result network” and write DM_subtypes_DIAMOnD in the box “Name of the result network”. A new network will emerge (Fig. 7), with the previously selected nodes (seeds) in yellow. To change the color of the primary nodes to differentiate them from the added ones, proceed as in point 3 from Section 4.
• 13.To save the image of the network, proceed as described in point 6 from Section 2.
• 14.To obtain the intersection of the two networks (two sets of disease module genes) using Cytoscape, go to "Tools" > "Merge" and click "Networks". Select the networks HF_subtypes_DIAMOnD and DM_subtypes_DIAMOnD in the "Available Networks" box and add them to the box "Networks to Merge". Then, select "Intersection" and click "Merge". The resultant Merged Network is represented in Fig. 8.
• 15.To save the image of the network, proceed as described in point 6 from Section 2.
• 16.Next, select all the genes from Fig. 8 by holding the Ctrl key, left-clicking, and selecting the network. Go to "Apps" > "NeDRex" > "Drug Prioritization" and click “Rank drugs with Closeness Centrality”. Then, select “Include only approved drugs” and rename the network as HF_DM_merged_DIAMOnD_Closeness Centrality (Fig. 9). Maintain the number of top-ranked drugs as 50. The primary seeds show up in yellow once more. Maintain the selection and bypass the color to yellow, for example.
• 17.To change the color of the drugs according to the drug group, proceed as mentioned in point 9 from Section 2.
• 18.To save the image of the network, proceed as described in point 6 from Section 2.
• 19.Next, you can apply the same workflow to retrieve new potential drugs with an original indication for treating obesity. To obtain the corresponding network, simply follow the instructions described for diabetes while selecting the disorder “mondo.0011122” instead (see Supplementary Fig. S1, Supplementary Fig. S2, Supplementary Fig. S3 obtained for this analysis). In the end, you will obtain the network presented in Fig. 10.

  Next, you can apply the same workflow to retrieve new potential drugs with an original indication for treating obesity. To obtain the corresponding network, simply follow the instructions described for diabetes while selecting the disorder mondo.0011122 instead (see Supplementary Figs. S1, S2, and S3 obtained for this analysis). In the end, you will obtain the network presented in Fig. 10.

Fig. 6.

Network of the genes associated with diabetes mellitus and other disorder subtypes, according to MONDO and NeDRex databases. Blue square nodes represent genes, and pink diamond ones correspond to the disorders. Databases sourced on January 9th, 2025. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 7.

Network of the disease module obtained using DIAMOnD for diabetes mellitus. Yellow nodes represent the genes previously selected, and the blue nodes represent the added ones. Databases sourced on January 9th, 2025. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 8.

Network resulting from the network intersection with the genes associated with heart failure after the DIAMOnD algorithm application and the network with the genes associated with diabetes mellitus after the DIAMOnD algorithm application. Databases sourced on January 9th, 2025.

Fig. 9.

Network of the drugs associated with the genes common to heart failure and diabetes mellitus after DIAMOnD algorithm application. Yellow square nodes represent the genes primarily selected, and the blue nodes are the ones added after the application of the DIAMOnD algorithm. Hexagon nodes correspond to drugs (green-approved; orange-approved, investigational; gray-other group drugs). Databases sourced on January 9th, 2025. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Supplementary Fig. S1.

Network of the genes associated with obesity and other disorder subtypes, according to MONDO and NeDRex databases. Blue square nodes represent genes, and pink diamond ones correspond to the disorders. Database sourced on January 9th, 2025

Supplementary Fig. S2.

Network of the disease module obtained using DIAMOnD for obesity. Yellow nodes represent the genes previously selected, and the blue nodes represent the added ones. Database sourced on January 9th, 2025

Supplementary Fig. S3.

Network resultant from the network intersection with the genes associated with heart failure after the DIAMOnD algorithm application and the network with the genes associated with obesity after the DIAMOnD algorithm application. Database sourced on January 9th, 2025

Fig. 10.

Network of the drugs associated with the genes common to heart failure and obesity. Yellow square nodes represent the genes primarily selected, and the blue nodes are the ones after the application of the DIAMOnD algorithm. Hexagon nodes correspond to drugs (green-approved; orange-approved, investigational; gray-other group drugs). Databases sourced on January 9th, 2025. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

5. Sex-specific drug discovery for heart failure using NeDRex

Sex differences can explain distinct responses to therapy, and a deeper knowledge of these differences can contribute to achieving sex-tailored treatment strategies [36,37]. To gain further insights into this issue, we conducted a non-exhaustive literature search for proteomic studies in HF involving men and women. Among these studies, Chandramouli C. et al. emerged as particularly noteworthy because it addressed sex differences in coronary microvascular dysfunction (CMD) in HFpEF [38]. Despite the recent advances in HFpEF treatment, comprehension of the different HFpEF phenotypes can help find more individualized and targeted therapies [39]. Coronary microvascular dysfunction has been pointed out as an important factor in HFpEF pathogenesis, in which several comorbidities, including obesity, hypertension, and diabetes mellitus, are suggested to contribute through the induction of a systemic pro-inflammatory state. Such comorbidities also induce oxidative stress, leading to dysregulation of the cGMP-PKG pathway in myocardial cells, contributing to cardiomyocyte hypertrophy and cardiac fibrosis [40]. However, little is known about sex differences in microvascular dysfunction. Recently, CMD was verified to be present in 85 % of an HFpEF population. Although its prevalence was found to be similar between sexes, 24 % of the patients presented endothelium-dependent microvascular spasms, and these were more often women [41]. This suggests that distinct mechanisms can be related to CMD in both sexes, underscoring that CMD should be more investigated in both HFpEF men and women, intending to find sex-specific therapeutic approaches.

According to the article [42], tumor necrosis factor receptor superfamily member 14 (TNFSF14), proteinase 3 (PRTN3), pregnancy-associated plasma protein A (PAPP-A), transglutaminase 2 (TGM2), B-type natriuretic peptide (BNP), chemokine ligand 20 (CCL20), interleukin-6 (IL6), phosphoinositide-3 kinase (PI3K), cluster of differentiation 93 (CD93), and epithelial cell adhesion molecule (EpCAM) were circulating proteins that were associated with CMD (coronary flow reserve < 2.5) in HFpEF men. In HFpEF women, the proteins associated with CMD were CUB domain-containing protein 1 (CDCP1), cathepsin L1 (CTSL1), decorin (DCN), delta/notch-like epidermal growth factor-related receptor (DNER), FMS-like tyrosine kinase 3 ligand (Flt3L), growth differentiation factor 15 (GDF15), gastric intrinsic factor (GIF), insulin-like growth factor-binding protein 1 (IGFBP1), insulin-like growth factor binding protein 7 (IGFBP7), junctional adhesion molecule A (JAM-A), metalloproteinase-9 (MMP9), prostasin (PRSS8), phage shock protein D (PSPD), spondin-1 (SPON1), tissue factor pathway inhibitor (TFPI), and tumor necrosis factor receptor superfamily member 13B (TNFSF13B). Gene names of these proteins were searched in UniProt (https://www.uniprot.org), and the resultant output was: TNFSF14, PRTN3, PAPPA, TGM2, NPPB, CCL20, IL6, PIK3CA, CD93, EPCAM, CDCP1, CTSL, DCN, DNER, FLT3LG, GDF15, CBLIF, IGFBP1, IGFBP7, F11R, MMP9, PRSS8, SFTPD, SPON1, TFPI, and TNFSF13B. Thus, to find therapies related to these proteins differently associated with CMD in men and women, we followed the following steps:

• 1.Select the Network_1 in the control panel of Cytoscape, click "Apps", select "NeDRex", and click “Quick Select”. In the “Choose the node type you want to select from:”, in the box “<Select type>”, click the arrow and “Protein”. In the box below “Type in the disease/drug/gene/protein/pathway name or ID:”, write the gene name of your protein of interest. For instance, write TNFSF14 and click on “Add to the list”. The name of the gene will appear in the box “Added”. Do the same for the other genes of interest in the case of men: PRTN3, PAPPA, TGM2, NPPB, CCL20, IL6, PIK3CA, CD93, and EPCAM. Verify if all your gene names are in the last box. If not, you can click “Remove last” to eliminate the last gene name added or click “Reset Selection” to eliminate all the gene names. If your list of gene names added is correct, then proceed, click “Select in Network”, and close the box. After this, the genes will be selected from the imported NeDRex Network.
• 2.Next, to expand your network, go to "Apps" > "NeDRex" and click on “MuST on current network”.
• 3.In the Algorithm settings, select the option “Return multiple Steiner trees” and keep the default parameters ("The number of Steiner trees" as 10, "Max number of iterations" as 5, "Penalize hub nodes" not selected). Rename the resulting network: Men proteins_MuST. You will obtain the network from Fig. 11A. The MuST algorithm generates a connected network that possibly includes the genes involved in the disease pathways and mechanisms by selecting genes associated with a disease of interest (seeds). The genes obtained after the application of this algorithm can be targets of putative repurposable drugs. To learn more about this algorithm, consult https://nedrex.net/tutorial/methods.html.
• 4.To save the image of the network, proceed as described in point 6 from Section 2.
• 5.Next, with all proteins from Fig. 11A selected, go to "Apps" > "NeDRex" > "Drug Prioritization" and click “All drugs targeting the selection”. Then, select “Include only approved drugs”, and rename the resulting network as Men proteins_MuST_All drugs (Fig. 11B). Change the color of drugs according to the drug group as mentioned before.
• 6.To save the image of the network, proceed as described in point 6 from Section 2.
• 7.Repeat the same workflow with the proteins found to be more relevant in women (see Supplementary Fig. S4A and B presenting the networks obtained in this analysis).

  Repeat the same workflow with the proteins found to be more relevant in women (see Supplementary Fig. S4A and B presenting the networks obtained in this analysis).

Fig. 11.

Protein-protein interaction network resulting from the application of the MuST algorithm to the proteins in yellow (proteins differently associated with CMD in men) (A). Network of the drugs associated with the proteins in yellow (proteins differently associated with CMD in men) and the blue ones added after MuST algorithm application (B). Blue nodes represent the proteins added after the algorithm application. Hexagon nodes correspond to drugs (green-approved; orange-approved, investigational and approved, experimental; gray-other group drugs). Databases sourced on January 9th, 2025. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Supplementary Fig. S4.

Protein-protein interaction network resultant from the application of the MuST algorithm to the proteins in yellow (proteins differently associated with CMD in women) (A). Network of the drugs associated with the proteins in yellow (proteins differently associated with CMD in women) and the blue ones added after MuST algorithm application (B). Blue nodes represent the proteins added after the algorithm application. Hexagon nodes correspond to drugs (green-approved; orange-approved, investigational and approved, experimental; gray-other group drugs). Database sourced on January 9th, 2025

In Cytoscape, the NeDRex application allows the representation of relationships (edges) between different nodes, like genes, proteins, diseases, and drugs, which helps identify drug-repurposing candidates. This kind of approach has the benefit of joining different types of data in a network. Also, it allows for insights into possible diseases and drug mechanisms. However, it does not permit us to see the drug's effect on a gene/protein. In this sense, bioinformatics tools, such as STITCH (Search Tool for Interactions of Chemicals), can be very useful. When using this tool in the context of a disease, it is important to consider the output as crude predictions requiring contextualization in the studied disease since it does not allow data to be filtered by disease. So, the utilization of more than one approach can be complementary to obtain the maximum information about our drug of interest.

In the context of drug repurposing, the next tools can help understand how a drug acts on a given protein in a pathway. Also, they can support the data found from the previous tool, that is, after finding a certain drug that can be used in a different pathological context, by giving insight into the signaling pathways/biological processes where such drugs can act.

Now, we will show bioinformatics tools that predict drug-target interactions using distinct approaches. For instance, SwissTargetPrediction can be included in a group of tools that predict drug-target interactions based on the similarity of drug structures, supported on the premise that similar compounds are likely to bind to the same target. This tool uses a combination of 2D and 3D structure similarity measures that, according to the developers, can improve target prediction accuracy if the query molecule is new or an outlier compared with the training compound series. Most of the tools that use a similar approach only base their predictions on the similarity of 2D structure. For example, the tools ChemProt 3.0 and SuperPred are two of the most cited but use only a 2D similarity-based approach [43,44]. STITCH uses a different approach to predict drug-target interactions that is based on biological networks. STITCH is a well-established resource with widespread use. Importantly, using the Cytoscape plugin stringApp, it is possible to import functional associations or physical interactions between protein-protein and protein-chemical pairs from different platforms, including STITCH. Other tools use a similar approach, such as PROMISCUOUS, but in this tool, the network consists of drugs, proteins (targets), and side effects as the nodes, with drug-side effect, drug-target, target-target and drug-drug interactions working as edges [44]. However, this is limited by the knowledge available on the side effects of the drugs. Also, contrary to STITCH, it is not possible to work with PROMISCUOUS directly in Cytoscape.

From our previous analysis with NeDRex, niclosamide emerged as one of the drugs. To identify the targets of niclosamide, we can use SwissTargetPrediction and STITCH. We will start explaining the operation with SwissTargetPrediction, which is particularly useful to this type of analysis, and then we will show how we can collect additional information with STITCH, such as the type of interaction between proteins and drugs.

6. Prediction of protein targets of drugs using the SwissTargetPrediction tool

The SwissTargetPrediction web tool allows one to predict the targets of a bioactive small molecule using a combination of 2D and 3D similarity measures to compare the query molecule to a library of active compounds on targets of 3 different organisms (Homo sapiens, Mus musculus, and Rattus norvegicus) [45]. An upside of this tool is the possibility of drawing any chemical structure, allowing the prediction of potential target proteins for new/experimental small molecular drugs. The proposed workflow for SwissTargetPrediction analysis is as follows:

• 1.On the SwissTargetPredicition website (http://www.swisstargetprediction.ch/), search for the option “Select a species” and select the one you are working with. For this tutorial, we selected Homo sapiens.
• 2.In the box “Paste a SMILES in this box, or draw a molecule”, paste a SMILES corresponding to niclosamide, for instance:

  C1=CC(=C(C=C1[N+](=O)[O-])Cl)NC(=O)C2=C(C=CC(=C2)Cl)O

  SMILES can be obtained in small molecules-specialized databases such as PubChem [46] or DrugBank [47].

• 3.Confirm that the molecular structure appearing in the draw box corresponds to the drug compound you are searching for.
• 4.Proceed with “Predict targets”.
• 5.
  After a few tens of seconds, a table with a list of the targets for your compound (predicted using the method described in [48]) is shown, along with a box with the structure of your query compound and a pie chart presenting the target's classes. Regarding the table, you will find:

  • a.
    A list of the target proteins, ranked by probability (first column).
  • b.
    Their respective UniProt IDs (column 3) and ChEMBL IDs (column 4) with links to UniProt [49] and ChEMBL [50] databases, where one can get specific information about the retrieved protein.
  • c.
    A green bar showing the probability of that protein being a target of the queried compound (assumed as bioactive) (column 6).
  • d.
    The molecules, active on the protein, whose 3D and 2D structures are similar to the structure of your query molecule (column 7). This is based on the assumption that two similar bioactive molecules are likely to share their targets. Therefore, for a query molecule, it identifies the most similar molecules among a set of known ligands. The predicted targets are those bound by the ligands displaying the highest similarity with the query molecule [45]. By clicking on the numbers in this column, you can see the ligands whose 2D or 3D structure is similar to the structure of your query compound.
  • e.
    The class of the predicted targets (column 5).
• 6.Lastly, by using the options available in the “Export results” item, you can save your data in different formats (PDF, Excel, and CSV) and print it (Supplementary Table S1).

  Lastly, by using the options available in the “Export results” item, you can save your data in different formats (PDF, Excel, and CSV) and print it (Supplementary Table S1).

7. Step-by-step analysis of protein-protein and protein-drug interactions using the STITCH tool

Drug targets can also be predicted using STITCH, a web tool that provides a network view of drug and protein associations based on direct (physical) and indirect (functional) drug-protein interactions. Furthermore, STITCH allows the visualization of protein-protein and drug-drug interactions [16]. So, the main additional utility of this tool is the possibility to visualize, in a single network, all drug-protein, drug-drug, and protein-protein interactions and access the type of interactions (when available) between the partners in a large list of organisms. Incorporating tissue specificity is a valuable aspect of STITCH analysis. The ability to choose specific tissues plays a pivotal role in assessing a drug's effects across various bodily regions. It is noteworthy that sex-related disparities in gene expression are tissue-specific, with a significant number of genes showing differential expression in tissues such as breast, adipose, and skeletal muscle; however, these patterns can vary depending on the diverse cell type populations within a given tissue [51]. Next, the workflow for STITCH analysis is presented:

• 1.On the STITCH website (http://stitch.embl.de/), click on “Search” to perform your quest.
• 2.In the search option “Item by name”, write the name of the drug/compound you are searching for. In this case, we are looking for niclosamide. Alternatively, you can open the “Chemical structure(s)” tab (menu at the left) and paste the identifier SMILES correspondent to niclosamide.
• 3.Choose the organism of interest (Homo sapiens in our case). If you opt for the “Chemical structure(s)” option, you will have to confirm the structure of your compound in the next step.
• 4.Press “Search”.
• 5.A new output will now open, showing a drug-protein interaction network (Fig. 12A). Nodes represent proteins or drugs, while edges represent their predicted or known associations. The rod-shaped nodes represent the drugs, and the circular ones represent the proteins. By clicking the nodes, information regarding the proteins and drugs can be obtained. In the same way, by clicking the interaction edges, you can obtain information about the interaction. You can see the evidence (Experimental/Biochemical Data, Association in Curated Databases, Co-Mentioned in PubMed Abstracts and Predicted Interactions) suggesting the functional links and the predictions for the type of specific actions. For instance, you can access the articles that support the link between niclosamide and the proteins shown in the network by clicking on the depicted interactions and then in the item “Co-Mentioned in PubMed Abstracts”. You will find 5 different ways to visualize your interactions in the “Viewers” tab (Network, Experiments, Databases, Text-mining, and Coexpression). However, the “Network” option is probably the most visually attractive.
• 6.
  Some parameters can be adjusted in your network. Go to the “Settings” tab to open the “Basic Settings” box:

  • a.
    In the item “settings for protein-chemical interactions”, you can choose to visualize interactions between chemicals or/and stereoisomers as separate compounds.
  • b.
    It is possible to choose which kind of interaction sources are used by STITCH with the option “active interaction sources:” (e.g., Text-mining, Experiments, and Databases). For a first analysis, it can be useful to be the least stringent possible about the choice of the interaction sources (that is, you should select all).
  • c.
    Additionally, you can choose the minimum score for the interaction and the maximum number of interactors to be shown. However, this strategy can progressively result in a greater number of false positives or irrelevant targets.
  • d.
    One of the most important settings is related to the meaning of the network edges. The item “meaning of network edges” allows you to choose the meaning of the network edges to be used in your analysis. The user has 4 options for visualization (evidence, confidence, molecular action, or binding affinity) that serve different purposes. By selecting “evidence”, you can see what kind and how many different data sources helped to build the network (of course, conditioned by the sort of interactions selected in step 5). By selecting “confidence”, you may see edges with different thicknesses according to the strength of the data collected. By choosing “molecular action”, you can see what kind of interaction is established between proteins and drugs (for instance, green means “activation”, and red represents “inhibition”). Finally, by opting for “binding affinity”, you can see edges with different widths according to the protein-chemical binding affinity. On the tab “Legend”, you can find information about the meaning of the chosen network edges.
• 7.Click on "UPDATE" to see the network after the selection of the desired parameters.
• 8.Scroll down for the “Advanced Settings” box. Here, it is possible to filter a human interaction network so that only the proteins believed to be present in a specified tissue are shown. STITCH integrates tissue-specific protein expression patterns from two data sources (TISSUES database or ExpressionAtlas). Before supplementing the network with tissue data, users can choose if they want to use data from TISSUES or Expression Atlas. You can search for the niclosamide's targets in the heart by selecting ExpressionAtlas and then heart – Human Protein Atlas (Fig. 12B).
• 9.Go to the “Analysis” tab. In this section, you will find the main network stats and the most relevant annotated biological processes, molecular functions, and cellular components. To export all biological processes associated with the proteins retrieved, simply download the TSV file on “Save/Export”.
• 10.You can also export your current network, for example, in the PNG format, in the “Tables/Exports” table.
• 11.Finally, you can add or remove interactions from your network by clicking on the tabs “+ More” or “- Less”, respectively.
• 12.The interaction of niclosamide with sex hormone receptors can also be explored using the STITCH tool by selecting “Multiple names” in the SEARCH, writing “niclosamide”, “estrogen receptor alpha”, “estrogen receptor beta”, “androgen receptor”, and “G protein-coupled estrogen receptor”, and selecting in “Advanced Settings” the ExpressionAtlas and heart – Human Protein Atlas (Fig. 12C).

Fig. 12.

Network depicting niclosamide interactions, using the STITCH tool (A). Networks depicting niclosamide interactions in the heart (selection of ExpressionAtlas and heart – Human Protein Atlas) (B). Network depicting niclosamide interactions with androgen receptor (AR), estrogen receptor alpha (ESR1), estrogen receptor beta (ESR2), and G protein-coupled estrogen receptor 1 (GPER) (C). Databases sourced on January 13th, 2025.

8. Discussion

Drugs typically used in HF treatment can be found in the network that resulted from the first analysis (Section 2, Fig. 2B). For instance, angiotensin I converting enzyme (ACE) inhibitors such as enalaprilat, captopril, moexipril, ramipril, trandolapril, fosinopril, and perindopril were found associated with the gene ACE. Other drugs usually used in HF are angiotensin II receptor blockers (ARB), such as azilsartan, candesartan, eprosartan, irbesartan, losartan, olmesartan, telmisartan, and valsartan. These drugs appear associated with the gene AGTR1 (angiotensin II receptor type 1). Associated with the genes ADRB1 (adrenoceptor beta 1) and ADRB3 (adrenoceptor beta 3), the network shows drugs such as carvedilol, bisoprolol, metoprolol, and nebivolol that are beta-adrenergic blockers, also used in HF treatment [52]. By searching for the pathways related to the genes associated with HF using the ShinyGO tool, an association between the renin-angiotensin-aldosterone system and the cGMP-PKG, cAMP, and calcium signaling pathways with the HF-related genes was shown, supporting their relevance in HF pathophysiology and their utilization as therapeutic targets [53].

It is interesting to note that some drugs with mechanisms of action that are not immediately associated with HF have also been proposed as candidates for repurposing. With the application of the DIAMOnD algorithm, new drugs emerged, such as niclosamide (an anthelmintic drug), associated with the gene LYN (LYN proto-oncogene, Src family tyrosine kinase). Indeed, a recent study showed that oral administration of niclosamide improved transverse aortic constriction-induced cardiac hypertrophy, cardiac fibrosis, and cardiac dysfunction in male mice, and the underlying mechanisms include enhancing mitochondrial respiration of cardiomyocytes, inhibiting collagen secretion from cardiac fibroblasts, and reducing the serum inflammatory mediator IL-6 [54]. Moreover, the effect of niclosamide on calcification in porcine aortic valve interstitial cells induced by a pro-calcifying medium was studied, and this drug was found to alleviate calcification, at least in part, by targeting the GSK-3β/β-catenin signaling pathway via inhibition of AKT and ERK [55]. The node associated with the niclosamide only appeared after feeding the network with the DIAMOnD algorithm, demonstrating its usefulness in identifying new potential drug/target pairs.

Among the drugs retrieved from the HF and diabetes intersection network, the drug niclosamide was also highlighted. This drug has been investigated in experimental studies in diabetes [56] and in the cardiovascular field [54,55]. The experimental validation of these associations, through in vitro studies and animal models of the disease, is crucial to evaluating their potential use in clinical trials. NeDRex can help us in this task by prioritizing drug targets for repurposing.

Niclosamide and perhexiline were only identified in the network derived from the analysis of proteins associated with microvascular endothelial dysfunction in women with HFpEF (Table 1, Supplementary Fig. S4). Perhexiline is highlighted as being associated with the protein EGFR. This drug is an inhibitor of carnitine palmitoyl transferase 1, which regulates the uptake of long-chain fatty acids into mitochondria [57]. Indeed, a previous study showed that this drug exerted significant benefits on V̇O₂max, left ventricle ejection fraction, symptomatic status, left ventricle function at rest and during peak stress, and skeletal muscle energy metabolism in patients with stable chronic HF [58]. Interestingly, belimumab, a drug mainly used for the treatment of resistant systemic lupus erythematosus in adults, is present in the network from Supplementary Fig. S4 [59]. Recently, it was published the first case reporting a 51-year-old Caucasian woman with HFpEF secondary to systemic lupus erythematosus treated successfully with belimumab (a monoclonal antibody that specifically inhibits the biological activity of soluble B-lymphocyte stimulator protein) supporting the idea that targeting inflammation can have a therapeutic potential in a particular group of HFpEF patients [60]. Importantly, the effects of the drugs obtained from the resultant analysis should be experimentally evaluated using in vitro models (in cells from male and female donors) and animal models (also representative of both sexes) before considering clinical applications.

Table 1.

Drugs associated with the proteins related to coronary microvascular dysfunction in HFpEF retrieved from NeDRex analysis exclusive from men and women, and common to both sexes.

Men	Women			Common
Artenimol	Acalabrutinib	Crizotinib	Necitumumab	Antithymocyte immunoglobulin (rabbit)
Becaplermin	Adenosine	Dabigatran etexilate	Nelfinavir	Bacitracin
Binimetinib	Adenosine phosphate	Dabrafenib	Neratinib	Carvedilol
Bortezomib	Afatinib	Dacarbazine	Niacin	Ceritinib
Brentuximab vedotin	Alteplase	Dacomitinib	Niclosamide	Cisplatin
Cabazitaxel	Aluminium	Dalteparin	Nilotinib	Copper
Caffeine	Aminocaproic acid	Deferoxamine	Ornithine	Dasatinib
Catumaxomab	Amodiaquine	Dexibuprofen	Osimertinib	Docetaxel
Colchicine	Anistreplase	Diiodohydroxyquinoline	Pacritinib	Foreskin keratinocyte (neonatal)
Dimethyl fumarate	Aprotinin	Dimercaprol	Palbociclib	Gefitinib
Duvelisib	Argatroban	Dobutamine	Panitumumab	Infigratinib
Eribulin	Astemizole	Dopamine	Pentamidine	Lenvatinib
Foreskin fibroblast (neonatal)	Axitinib	Ebastine	Perhexiline	Midostaurin
Glucosamine	Beclomethasone dipropionate	Econazole	Prazosin	Natalizumab
Heparin	Belimumab	Edetic acid	Regorafenib	Nintedanib
Hexylresorcinol	Bithionol	Empagliflozin	Reteplase	Pazopanib
Hypromellose	Boceprevir	Encorafenib	Rivaroxaban	Ponatinib
Idelalisib	Bosutinib	Erlotinib	Secobarbital	Sunitinib
Ixabepilone	Budesonide	Florbetaben (18F)	Selumetinib	Tirbanibulin
Lauric acid	Busulfan	Florbetapir (18F)	Sodium aurothiomalate	Vandetanib
Nimesulide	Cabozantinib	Flucloxacillin	Sorafenib	Zinc
Ocriplasmin	Calcium citrate	Fluphenazine	Streptokinase
Oxymetholone	Calcium Phosphate	Flutemetamol (18F)	Tacrine
Paclitaxel	Candesartan cilexetil	Gentian violet cation	Tamoxifen
Parecoxib	Captopril	Glutathione	Telaprevir
Podofilox	Cefalotin	Hexachlorophene	Tenecteplase
Prednisolone	Cefonicid	Hexamidine	Terfenadine
Raloxifene	Cefotetan	Human C1-esterase inhibitor	Thioridazine
Siltuximab	Cefotiam	Ibrutinib	Tivozanib
Technetium Tc-99m nofetumomab merpentan	Cefradine	Iloprost	Trabectedin
Trastuzumab emtansine	Cefuroxime	Imatinib	Tranexamic acid
Vinblastine	Cetuximab	Insulin human	Tretinoin
Vincristine	Chlorhexidine	Lapatinib	Tromethamine
Vinflunine	Chloroquine	Levodopa	Tucatinib
Vinorelbine	Chlorpromazine	Lidocaine	Urokinase
	Cholecystokinin	Lonafarnib	Vemurafenib
	Chromic chloride	Lysine	Vitamin A
	Citric acid	Methyldopa	Vorinostat
	Clioquinol	Methylene blue	Ximelagatran
	Clomifene	Miconazole	Zafirlukast
	Clotrimazole	Minocycline	Zoledronic acid
	Coagulation factor VIIa Recombinant Human	Mitoxantrone
	Colistin	Montelukast
	Conestat alfa	Mupirocin

From the analysis with the SwissTargetPrediction tool, the retrieved protein targets of niclosamide include STAT3 (Signal transducer and activator of transcription 3) and KCNMA1 (Calcium-activated potassium channel subunit alpha-1). STITCH shows that the protein STAT3 is a target of niclosamide. Additionally, this tool gave us a clue about the type of action that niclosamide exerts on this protein, showing that niclosamide binds STAT3. Also, niclosamide inhibits the protein TNF (tumor necrosis factor). Thus, this kind of information can help us understand the mechanism of action for our query compound and help guide us in the design of validation assays. For instance, herein, we could validate niclosamide's therapeutic activity by using a STAT3-interacting protein to block their interaction. Niclosamide's targets can be tissue-specific, as observed in the analysis with STITCH showing that niclosamide inhibits the protein NOTCH1 (Neurogenic locus notch homolog protein 1) in the heart. The distribution of target proteins among different tissues is a determinant factor for the drug's impact on the organism and its efficacy [61]. Curiously, niclosamide might interact indirectly with the androgen receptor (AR) and estrogen receptor alpha (ESR1) through the protein STAT3 in the heart. This can help in the design of experimental approaches intending to validate and explore specific drug-protein interactions.

9. Conclusion

Herein, we used the HF syndrome to show the usefulness of the Cytoscape's application, NeDRex, in finding candidate drugs to be repurposed in this condition. ShinyGO was also used to perform a gene enrichment analysis to gain mechanistic insight into our generated gene dataset. Moreover, the drug niclosamide, uncovered by NeDRex analysis, was used to demonstrate the applicability of the bioinformatics tools SwissTargetPrediction and STITCH in the prediction of the targets and protein interactions of a molecular drug, respectively. The approach described in this article can also be very useful in the identification of potential sex-specific drugs and interactions between drugs and sex hormone receptors.

Our analysis with NeDRex considering circulating proteins associated with coronary endothelium remodeling in men and women with HFpEF allowed the identification of distinct drug targets in men and women. SwissTargetPrediction allowed us to search for the protein targets of niclosamide and showed that the retrieved targets include, for example, STAT3. STITCH was used to elucidate niclosamide's mechanism of action, showing that this drug binds STAT3. Additionally, its tissue selection option can greatly help determine the drug's impact on different tissues.

The dynamic nature of biological systems cannot be dismissed, and the fact that molecular association networks can only translate specific points in time demands a critical analysis of the results obtained with the utilization of bioinformatics tools. The combination of a computational and an experimental approach is thus important to first identify and then validate a drug as a candidate for repurposing.

Importantly, the analysis presented in this tutorial, as well as the dataset used from a specific article, were only selected as examples to demonstrate the usefulness and practicability of the bioinformatics tools explored, and thus, no biological meaning should be attributed to the results obtained.

The following are the supplementary data related to this article.

Supplementary Table S1

Protein targets of niclosamide, retrieved using the SwissTargetPrediction tool. Database sourced on April 9th, 2025

Supplementary material

CRediT authorship contribution statement

Ivo Fonseca: Writing – original draft. Fábio Trindade: Writing – review & editing. Mário Santos: Writing – review & editing. Adelino Leite-Moreira: Writing – review & editing. Daniel Moreira-Gonçalves: Writing – review & editing. Rui Vitorino: Writing – review & editing, Conceptualization. Rita Ferreira: Writing – review & editing, Conceptualization. Rita Nogueira-Ferreira: Writing – review & editing, Supervision, Formal analysis, Conceptualization.

Funding

R.N.-F. acknowledges FCT for the research contract CEECIND/03935/2021 (https://doi.org/10.54499/2021.03935.CEECIND/CP1685/CT0001) under the CEEC Individual 2021.

Declaration of competing interest

The authors declare there are no conflicts of interest.