Getting started in tellic graph

Three steps to begin your biomedical exploration in tellic graph

tellic graph is optimized for Google Chrome. Please use Chrome when using tellic graph.

Having spent endless hours on traditional manual literature reviews, you are now ready to add automated knowledge gathering to your workflows. Here we show how tellic graph makes it easy to quickly surface actionable and meaningful insights in three easy steps: (1) enter search terms, (2) connect the terms at a click of a button, and (3) review results.

Step 1 – Search 

All research starts with entering your search terms onto the blank web interface. Being able to easily search for entity-centric content is critical to receiving relevant results that advance your research. tellic graph makes searching for entities of interest easy with synonym matching. As you type in the search term, the tool uses various pieces of metadata such as entity names, ontology IDs, and synonyms to bring up the list of entity records — including all synonyms found in literature — from which to select. Choose your record of interest by clicking on it, and the entity will appear on the interface (Figure 1).

image-4

Continue to add more entities as you like - the real power of graph technology is in finding connections among many entities, so the sky is the limit here. To keep your graph organized, arrange the entities by grabbing and moving them.  You can add more entities later, but now you are ready to connect.

Step 2 - Connect  

The next step after adding entities of interest to your knowledge graph is expanding and exploring to connect the dots and identify relationships of interest. While graph data does provide the opportunity to explore millions of potential biomedical relationships (tellic graph alone has over 200M+ relationships to explore), it is critical to have tools and controls in place to keep your research focused on addressing your subject or hypothesis. Here we will look at two approaches to connect the dots in graph: entity exploration and hypothesis exploration.

Entity exploration refers to the organic exploration of entities of interest by looking at known relationships and filtering the results based on key, science-driven criteria. For example, when looking broadly at all relationships to COVID-19 (see Figure 2a), it is difficult to quickly identify any actionable next steps for your research despite the large amounts of relationships. While AI has completed the task of identifying all known relationships, there is still effort needed to prune down and filter the graph to find the most relevant knowledge. 

To quickly explore large graphs, apply filter criteria based on research goals. In the COVID-19 example (see Figure 2b), exploration is focused on relationships between small molecule drugs and COVID-19 with an average strength greater than 85%. Entity exploration is an approach we frequently see applied by scientists looking to “catch-up” on a subject-area when rolling onto a new project. Another use case is to follow up on publications or data, for example, to explore a potential target or disease of interest for a given R&D program.  

fig2

Hypothesis exploration refers to testing or validating specific hypotheses in graph to evaluate the strength or supporting data quality for a particular relationship. This approach generally starts with two or more entities and involves exploring direct relationships between them. If no direct relationships exist, exploring intermediate relationships or mutually related entities is an effective way to continue exploration. In Figure 3 we illustrate testing relationships between COVID-19 and three entities of interest: ACE2 (gene), Remdesivir (drug) and rs1427407 (variant).  

fig3

Selecting all four nodes and clicking explore shows direct relationships with ACE2 and Remdesivir, but no direct relationships with rs1427407. Based on the visualization we can also see that the relationship between COVID-19 and ACE2 has a stronger relationship (denoted by the 85% edge label) as well as more supporting research (denoted by the relative thickness of the edge) as compared to the relationship between COVID-19 and Remdesivir. However, exploring direct relationships does not provide any insight into how rs1427407 might be related to COVID-19.  

fig4

To further this exploration between COVID-19 and rs1427407, we can leverage “Find mutually related nodes” functionality within tellic graph. This feature, unique to graph data, returns all mutually related entities for selected nodes of interest. As we can see in Figure 4, both COVID-19 and rs1427407 share a relationship with Hemoglobin subunits. Review of document-level metadata can then inform on the relevance of the relationships. Without access to tellic graph, a hidden mutual relationship such as the one between Hemoglobin subunits, COVID-19, and rs1427407 would take hours of research to surface, if at all. This provides clear, actionable next steps for follow-up review of supporting data to explore available knowledge around these entities of interest. 

Step 3 – Review & Action  

While identifying relationships between entities provides a basic insight, it still lacks actionable, data-driven context. The final step is to drill down into the underlying data to review supporting research and take action on the findings.  

fig5

As we can see in Figure 5, tellic graph provides detailed, document-level metadata on relationships of interest. For the example of COVID-19 and ACE2, we can see supporting research broken down to the sentence level, each with an associated Assertion Strength, source identification, journal information and direct link to the source document.  

Using this supporting research and accompanying metadata you can quickly identify relevant relationships and research further the most relevant ones.