Linked Discoveries^™ User Guide

Getting Started: Explore Linked Discoveries

To explore the research neighborhood surrounding a PubMed abstract, click the “Linked Discoveries” button, found in the right-hand column of abstract pages on larger screens and below the title on smaller screens. The Linked Discoveries button is only available on abstract pages for articles that represent original research, reviews, and related evidence.

Linked Discoveries opens in a new tab and shows a graph view of a research neighborhood surrounding the abstract you started from in PubMed (called a seed article in Linked Discoveries). The toggle above the graph can be used to see a timeline view of the neighborhood instead of the graph.

In Linked Discoveries, you can:

1. View the seed article. The seed article appears as the cyan node labeled “Seed” in the top left corner of the visualization. In the timeline view, a dotted blue line marks the publication year of the seed article.
2. View the neighborhood of related articles. Each article in the neighborhood is represented by a node labeled with the first author and year of publication. The color of the node shares information about the article type. See the key at the bottom of the visualization to identify article types.
3. Filter the neighborhood by shared topics. Filter by condition, gene, or chemical to narrow the neighborhood to articles sharing those topics. You can select one or more topics to filter on and choose whether the neighborhood is narrowed to records tagged with any of the selected topics, or all the selected topics.
4. View details for a selected node. By default, details for the seed article are shown in the article information card to the right of the graph or timeline. Click a different node to view its details instead. Filter the graph or timeline by clicking on the name of a specific condition, gene, or chemical on the information card. Click “View in table” to view the neighborhood in a structured table format.
5. Build a new neighborhood from one of the nodes. Click “Set as seed article” to generate a new neighborhood from any of the nodes in the current visualization.

Features in Depth

What is a Linked Discoveries Neighborhood?

A Linked Discoveries neighborhood is a network of related articles generated around a specific PubMed record (the seed article).

PubMed Records Included in Linked Discoveries

Linked Discoveries is only available for PubMed records representing original research, reviews, and related evidence. New PubMed records are added to Linked Discoveries daily.

Some PubMed content, such as biographies, interviews, and preprint records do not meet the criteria for inclusion in Linked Discoveries. Records without abstract text are also excluded.

PubMed records for publication updates like errata, expressions of concern, comments, and notices of retraction cannot be a seed article; however, these updates are listed on the article information card for the articles that they are linked to so that you can identify any information relevant to an article’s standing in the literature.

How a Neighborhood is Created

Linked Discoveries uses a version of BiomedBERT (previously called PubMedBERT), an open-source language model trained specifically on biomedical and clinical text, that has been fine-tuned to judge how closely two pieces of text are related to create a neighborhood of related articles. The model determines whether the topics in two PubMed abstracts are related to the other by comparing the text from the titles, abstracts, and keywords. For more information about neighborhood generation, see Neighborhood Technical Information.

Articles in the graph view are arranged in order of similarity to the seed article, read left to right and top to bottom. Linked Discoveries draws from a very large, densely connected body of literature, meaning similarity scores within a neighborhood are often extremely close to one another and small differences in score or order should not be taken to mean that one article is substantially more relevant than another. The graph view is intended to portray a group of closely related articles rather than imply a strict similarity ranking.

You can change the size of the neighborhood using the display options above the graph to narrow the neighborhood to the most related records or expand the neighborhood to include up to 200 records. As the size of the neighborhood increases, new nodes are added to the lower right corner of the graph view.

Reading the Graph and Timeline

Linked Discoveries offers two ways to view a research neighborhood:

• Graph view arranges the nodes as a connected network, making it easier to see how articles relate to one another through shared topics and citations.
• Timeline view arranges the same nodes by publication year, making it easier to see how research within the neighborhood has developed over time.

The information available for each node remains the same across both views:

• Each node (circle) represents one PubMed record. The nodes are labeled with the first author’s name and publication year.
• Node color provides more information about article type and publication updates, and a node may display more than one attribute at a time. The key below the graph or timeline indicates the meaning behind the colors:
  • Seed article: The cyan node labeled “Seed” shows the PubMed record the neighborhood is built around. This may be the abstract you started from in PubMed, or a node you selected as a new seed from a previous Linked Discoveries neighborhood.
  • Article: Dark blue nodes show other articles in the neighborhood.
  • Retracted: A red node indicates that an article has been retracted. A link to the retraction notice in PubMed is available in the article information card.
  • NIH funded: Light blue on the right side of a node indicates that an article has been supported by funding from one or more of the 27 Institutes and Centers of the National Institutes of Health. Funding labels are generated from funding data available in PubMed.
  • Review: Gold in the lower half of a node indicates that it is a review article. Review articles examine findings from multiple studies to summarize or evaluate what is currently known about a topic and include meta-analyses, scoping reviews, systematic reviews, guidelines, and consensus statements. Reviews are identified using the Medical Subject Heading (MeSH) publication types indexed on the PubMed record.
  • Selected article: A bright blue circle around a node shows the article you currently have selected in the visualization.
• Halo size shows the citation count for an article. A bigger halo means the article has been cited more frequently by other articles in the neighborhood, based on data available in PubMed.
• Citation links are shown by lines connecting nodes in the neighborhood. See Citation relationships for more information about interpreting citation links.
• Circular icons displayed near the top of nodes show shared condition, gene, and chemical topics when one or more topics is selected from the filter menu.

Shared Topics in the Neighborhood

Linked Discoveries surfaces shared condition, gene, and chemical topics using the National Center for Biotechnology’s (NCBI) Entrez^® link system, which connects PubMed records to related resources across NCBI databases.

• Condition topics come from PubMed records linked to MedGen, a database of genetic and medical conditions.
• Gene topics come from PubMed records that are cited in Gene, a database that aggregates gene-specific information from RefSeq, model organism databases, and other resources.
• Chemical topics come from PubMed records that are cited by compound records in PubChem, a database that aggregates chemical and drug information contributed by data sources from around the world.

You can view the source record for any condition, gene, or chemical you see in Linked Discoveries by clicking the “View in [database]” link in the article information card to the right of the graph or timeline.

Filtering by Topic

You can narrow a neighborhood to nodes that share specific condition, gene, or chemical topics with the seed article using the filters above the visualization.

How to filter by topic

Three dropdown menus above the visualization allow you to filter the neighborhood by condition, gene, and chemical. Selecting a topic from any of these filters narrows the neighborhood to nodes sharing that topic.

You can select multiple topics at once across each of the three filters. Some topics are arranged hierarchically, with broader topics containing narrower ones. Selecting a broader topic automatically includes any narrower topics that are also present in the neighborhood.

When you select a filter, a circle icon is added above nodes that match that topic. The color of the icon indicates the type of match, with red icons indicating condition topics from MedGen, green icons indicating gene topics from Gene, and teal icons indicating chemical topics from PubChem.

Choose Any or All to control how your selections are combined:

• Any shows nodes that match at least one of your selected filters
• All shows nodes that match every one of your selected filters

The “Active filters” line above the neighborhood shows the currently applied filters. Filtering does not change the position of matching nodes; non-matching nodes are simply hidden from view.

Removing filters

To remove a filter, you can:

• Click the X icon on the filter you would like to remove
• Uncheck the box next to the topic name in the filter menu, or
• Click “Clear all” to remove all applied filters

Topic Data Availability

Linked Discoveries shows the condition, gene, and chemical topics indexed on PubMed records. Topic coverage may vary depending on available data, and the absence of a topic label on a PubMed record does not necessarily mean the article does not discuss the topic. Topic indexing may reflect the text and keywords found in the abstract, as well as standardized terminology and related concepts identified through the source database, so a tag may appear on a PubMed record even if that exact term isn’t used in the abstract itself.

Citation Relationships

Linked Discoveries shows two kinds of citation relationships between records in the neighborhood: Cited by and Cites (also called References in PubMed). You can view these relationships for all nodes in the neighborhood by selecting “Show All citation connections”. When this feature is selected, light gray lines connect records that cite each other throughout the neighborhood.

When you select a specific node in the graph or timeline,

• Cited by connections identify articles that cite the selected node. A solid blue line shows which other records in the neighborhood cite the selected node.
• Cites (references) connections identify articles that the selected node cites. A dotted blue line shows which records in the neighborhood are cited by the selected node.
• When “Show All citation connections” in the top right corner of the visualization is selected, solid gray lines connect papers that cite each other throughout the neighborhood.

Please note, “Cited by” and “Cites” relationships are shown when the citation data is available in PubMed. Citation links are generated using data submitted by publishers and from NCBI resources, when available. PubMed and Linked Discoveries may not include the complete list of works citing a particular item.

Article Information Card

You can find details about a specific article in the information card to the right of the neighborhood visualization. When the data is available, the information card includes:

• Article type icons used in the Linked Discoveries visualization to show retracted, NIH funded, and review articles when applicable.
• Citation data including the PMID, title, authors, and publication date. The author list is truncated to the first six author names followed by et al. Click on the PubMed ID at the top of the card to view the record in PubMed and access the complete author list.
• Publication Updates shares any updates to the selected article including retractions, errata, expressions of concern, and updates. Click on the link for a category to view the relevant record(s) in PubMed. When an article has been retracted, a red banner indicating the retraction also appears at the top of the information card.
• Abstract text for the selected node is available by expanding this section.
• Shared topics including conditions, genes, and chemicals that are linked to the selected node. Select the topic name on the information card to filter to that topic in the visualization. Select “View all in [database]” to view the related NCBI database records for the topic(s) indexed on the selected article.
• Citation relationships when citation data is available. Click “Cited by” to view the list of articles citing the selected record. Click “Cites” to view the list of articles the selected record cited.
• Click “Set as seed article” to build a new neighborhood around the currently selected node. The new neighborhood will open in a new tab.
• Click “View in table” to be taken directly to the neighborhood information in a structured table format.

Article Details Table

The data represented in the graph and timeline is also available in a table located below the visualization. The table includes one row for each article and columns including:

• Article Details including PMID, author list, journal, publication year, and if applicable, icons indicating article type and a red banner indicating if the article has been retracted
• Abstract text when available from PubMed
• Topics including the condition, gene, and chemical topics indexed on each record

More information about the table:

• The seed article appears at the top of the table. If a node other than the seed article is selected in the graph or timeline, a blue box outlines the record for that article in the table.
• The table is sorted by similarity to the seed article by default. You can use the dropdown menu above the table to sort by publication year instead.
• As you filter the neighborhood to specific condition, gene, and chemical topics, the table updates to match by removing any records that do not match the filter criteria.
• Author lists are truncated to the first 100 author names, followed by et al. Find the full author list in PubMed.

Save Your Results to a File

Use the “Download CSV” button above the table to save the currently displayed neighborhood to a file.

The file includes one row per article, with citation details (PMID, title, authors, journal, publication year, article type, and publication updates), the abstract, tagged condition/gene/chemical topics, and Cited by/Cites relationships.

More information about saving to a file:

• When filters are applied, the information saved to the file reflects only the currently visible records matching the filter criteria.
• Author lists are truncated to the first 100 author names, followed by et al. Find the full author list in PubMed.
• Publication updates are grouped by category (such as retractions, expressions of concern, errata, updates, and comments), each with a link to the corresponding records in PubMed.

View all in PubMed

At any point, click the “View all in PubMed” button located above the table to see the neighborhood citations in PubMed. When filters are applied in Linked Discoveries, the list of citations exported to PubMed is limited to the nodes matching the filter criteria.

From there, explore the neighborhood using PubMed tools such as:

• Sort and filter options to reorder the neighborhood by publication date, journal, or other criteria, or narrow the neighborhood down further using PubMed filters
• Results by Year Timeline to see the neighborhood publication history at a glance or download it as a CSV
• Search history to combine the neighborhood citations with other PubMed searches
• My NCBI Collections to save the neighborhood citations permanently to your My NCBI account for future reference
• Save to citation manager to export the neighborhood citations for use in your citation management software of choice

Provide Feedback and Request Help

Please reach out to the NLM Help Desk to request help with any questions.

Linked Discoveries is a prototype resource offered as a part of an ongoing effort to make it easier for users to identify connections across research findings. We welcome your feedback to help inform future directions of this resource. Use the yellow button on the right side of any Linked Discoveries page to share your feedback.

Appendix

Neighborhood Technical Information

Linked Discoveries compares PubMed records based on semantic meaning rather than keyword matching, using a language model called BiomedBERT. The text of each eligible record, including article title, abstract, and keywords, is converted into numerical representations of the content called embeddings. When compared, these embeddings enable Linked Discoveries to identify similar records even if they use different terms to describe similar concepts. Because this comparison relies on the abstract text, a record without an abstract cannot be used as a seed article or included in a neighborhood.

To generate a neighborhood around a seed article, Linked Discoveries retrieves PubMed records with the closest similarity scores (not shown), based on the precalculated embeddings. By default, the displayed neighborhood includes the seed article and the 49 most similar records. You can expand the neighborhood up to 200 nodes. Condition, gene, chemical, citation relationship, article type, and funding information are used to annotate the neighborhood after it is generated; this data is not used to create the neighborhood.

Linked Discoveries draws from a very large, densely connected body of literature, meaning similarity scores within a neighborhood are often extremely close to one another and small differences in score or order should not be taken to mean that one article is substantially more relevant than another. Similarity in Linked Discoveries should be treated as a starting point for exploration. Two articles can be closely related in meaning while reaching different conclusions or using different methods.