Explore biological networks illustrating how SARS-CoV-2 might affect its host cell. Built on evidence from the biomedical literature, networks are constructed using a large-scale Knowledge Graph and Machine Learning to predict gene and drug effects on selected biological processes, diseases, or pathways.

Scientific background

Recently Gordon et al^[1] identified human host proteins that interact with SARS-CoV-2 viral proteins using an affinity-purification mass spectrometry screen. These host proteins were functionally characterized and screened for existing drug targets to identify drugs that could potentially be repurposed against COVID-19.^[1] Building on this work, we connected the SARS-CoV-2 host protein interactions to biological functions or diseases affected by viral infection in a large-scale knowledge graph derived from the biomedical literature.^[2] Our aim is to explore how the virus could interfere with various host cell functions, and also to identify additional drug targets and other genes that could potentially be modulated against COVID-19. Our results are presented in the form of interactive network visualizations, that allow exploration of underlying experimental evidence, made available to the scientific community in the hope that they are useful for COVID-19 research.

What can I do on this site?

• Explore networks depicting how the coronavirus affects its host
• Display patterns of predicted activity of genes in these conditions
• Identify potential targetable gene by drug predicted to impact the effect of the virus

What is a network?

The networks on this site are hypotheses depicting how SARS-CoV-2 proteins interact with host proteins^[1] and the downstream effect those interactions may have on a biological function, disease, or pathway (i.e. the "outcome" of the network).^[2] The goal is to identify different mechanisms through which the virus may impact its host, and through that insight, reveal potential ways to stop it.

The shape of each node indicates the type of that entity (e.g. cytokine, enzyme, canonical pathway, etc.). The "Show Legend" link will display a legend that lists all the possible entity types and their shapes.

Colors overlaid on the network nodes indicate the predicted effect those molecules have on the outcome in two different scenarios you may choose:

• "promote" - the activation of orange nodes and inhibition of blue are predicted to promote the outcome
• "suppress" - the activation of orange nodes and inhibition of blue are predicted to suppress the outcome

In all scenarios, white nodes do not have a prediction but are involved in additional key interactions relevant to the network.

Pink borders around nodes identify host genes targeted by existing drugs.

How were these networks created?

The networks were created computationally using a Machine Learning model trained on the QIAGEN Knowledge Graph, a network of hundreds of thousands of entities connected through millions of findings curated from biomedical literature by subject matter experts over the last 20+ years.^[2]

After training, the model was used to predict and score genes that have an activating or inhibiting effect on a selected biological function, disease, or pathway. These genes were then connected within the Knowledge Graph to the interactome involving SARS-CoV-2 viral proteins and host genes recently documented by Gordon, et al.^[1] The connections were made through short directed paths from the viral proteins to the selected function, disease, or pathway, and based on a scoring function a subset was used to construct a network. As a consequence of this approach, in every network there will always be a causal path from any viral protein to the outcome through edges of curated findings that (a) is as short as possible, and (b) contains genes predicted with highest confidence to affect the outcome.

Finally, drug targets within each network were identified, and the effect of each drug on the outcome were computed, once again leveraging information from the Knowledge Graph.

What do the controls do?

• Select Network: Select which network to view from over 50 different options. Networks are labeled according to the biological function, disease, or pathway impacted by the virus through the relationships shown in the network. If any genes, drugs, or viral proteins are specified in the "Filter by" field, the list of networks will be limited to those containing those entities.
• Filter by: Optionally enter one or more genes, drugs, or viral proteins. The list of networks will be limited to those containing one or more of those entities. Additionally, the count of networks containing all of the entities will be updated in the label immediately below the control.
• Switch to promote/suppress: Click this link to show the function, disease, or pathway in a promoted/suppressed state and the pattern of host gene activity that is predicted to lead to that state. The activation of orange nodes and the inhibition of blue are predicted to result in that outcome.
• Show existing drug targets: Select the checkbox to identify host genes that can be targeted by known drugs. Drug targets will be highlighted with a pink border.
• Drag a node in the network to customize the layout.
• Click a node in the network to view more details about it in the upper right panel. When available, a link to the IPA Gene View page leads to additional information about the molecule. If one or more drugs target that gene, the drugs will be listed in the lower right panel. Each drug will be displayed with its name, effect on gene target, and predicted effect on the function, disease, or pathway through relationships in the network. Click on the drug name to open the IPA Chem View for that molecule.
• Click an edge (a relationship line) in the network to view an explanation of the relationship between the connected nodes. One finding will be displayed, along with a link to IPA to view all of the available findings supporting that relationship.
• Show Legend: Display a legend listing all the node shapes that appear in the networks and their meaning.

What is the purpose of changing promotion/suppression of a function/disease/pathway within a single network?

Why would I want to promote "Viral infection?"

Often the desired direction of a function, disease, or pathway can be different depending on your goal. For example, when evaluating how a virus attacks a cell, the function "Viral infection" will likely be promoted, but if considering a host's normal response or the pattern of gene activity during treatment, it may be better to view the same network as though the function were suppressed. Therefore, this option provides a way to look at the same mechanism from different perspectives. Additionally, the effect of a viral protein on a host gene is not always known, so this provides flexibility for you to choose the downstream effect based on your own understanding.

Is every host protein from the Gordon paper^[1] in a network?

No. Some of the host proteins mentioned in the paper are not known to bind with other molecules in the QIAGEN Knowledge Graph that are also covered in the Machine Learning model, so those molecules do not appear in these networks.

Is every viral protein from the Gordon paper^[1] in a network?

No. Only a subset of representative viral proteins mentioned in the paper have been used in these networks.

How may I use these hypothesis networks?

On this site, all networks, descriptions, and other information are freely available to everyone. Each network has a unique URL, so specific networks can be linked directly. If you would like to reference a network in your own work, please cite this project in the following way:

This network was provided by the Coronavirus Network Explorer created by the Project Insight team at QIAGEN, Inc. ({{baseUrl}}).

Can my data be incorporated into these networks?

The best way to evaluate your data is in our IPA application. In that software, you can upload expression, phosphorylation, and other forms of data; overlay it on pathways; analyze it to identify key canonical pathways, upstream regulators, and biological functions; and compare the results to over 60,000 curated analyses to find significant commonalities and differences. Click here to sign up for a QIAGEN IPA trial: https://digitalinsights.qiagen.com/products-overview/discovery-insights-portfolio/analysis-and-visualization/qiagen-ipa.

Can new networks for different functions/diseases/pathways be generated? Are you open to new collaborations?

To request new sets of networks or to collaborate on a research project, please reach out to us at insight@qiagen.com.

Our hope is that these networks provide some insight into how the virus operates and maybe even inspire a few ideas for further research.

^[1] Gordon, D.E., Jang, G.M., Bouhaddou, M. et al. A SARS-CoV-2 protein interaction map reveals targets for drug repurposing. Nature (2020). https://doi.org/10.1038/s41586-020-2286-9

^[2] Krämer A, Green J, Pollard J Jr, Tugendreich S. Causal analysis approaches in Ingenuity Pathway Analysis. Bioinformatics (2014) Feb 15;30(4):523-30. https://doi.org/10.1093/bioinformatics/btt703