PubGraph Query
  Basic Query
  Advanced Query
  Choosing Colors
Gallery -- Sample Output

PubGraph: visualizing results from PubMed

PubGraph is a visual interface to PubMed -- a way to get perspective on the biomedical research literature.

Users provide a set of PubMed queries as input. PubGraph summarizes the associations in the results from PubMed as a graph:
Different association measures are displayed in the graph, permitting patterns in the literature to be seen.


To use PubGraph, please go to the Basic Query interface and provide a set of PubMed queries that look like this:

In these example queries, [tiab] (title and abstract) [MH] (keyword, from the MeSH list of keywords) are PubMed search field tags. These tell PubMed to search for documents in specific ways. The PubMed online help gives more ideas about how to develop powerful PubMed queries. By choosing queries well we can get better perspective.
(This is an inactive example window. To use PubGraph, please enter queries on the Basic Query page.)
Each of these five lines in this text area specifies a single PubMed query. You can specify as many queries as you like, up to 100; PubGraph limits the number of queries to 100 since beyond that the amount of information usually obscures any pattern -- the trees become a forest.


PubGraph asks PubMed to find all relevant publications for these queries, and displays a result page that includes a graph representation:
association graph Each node of this graph represents a PubMed query.

Each edge between nodes that have queries X and Y represents the relationship between X and Y.

Each node is labeled with a number in parentheses -- this is the number of hits (publications) that PubMed has found for that query. So for example PubMed finds 59645 publications with the keyword dopamine.

Each edge is labeled with two numbers. If the nodes have queries X and Y, the top number is the number of hits for (X AND Y), and the bottom number is the number of hits for (X OR Y). The ratio of these numbers is called the Jaccard co-occurrence index; it measures the degree to which the two queries coincide among all publications.

For example, among the 59645 publications about dopamine and the 66148 publications about stress, 492 are about (dopamine AND stress), and 125301 are about (dopamine OR stress). The Jaccard index is 492/125301 = 0.0039 = exp(-5.5).
The result page also includes a heatmap, showing hierarchical clusters of queries reflecting their degree of association:
  heamap We can see that the "sleep disorders" and "stress, psychological" queries are clustered; their results have high Jaccard coefficient, which is a commonly-used measure of co-occurrence in a set of documents.

PubGraph also provides the results in a similar, but clickable, table:
Notice that the entry in the first row for the queries "dopamine [MH]" and "stress, psychological [MH]" are 492, 125301, and exp(-5.5) -- as explained above.

Some more examples showing how to use PubGraph are in the Gallery.

Tailoring the Output

PubGraph has many features for tailoring the output to provide different perspectives.

For example you can produce graphs with alternative colors for the nodes. This is done by optionally start each line with a Group ID, a single digit 0, 1, ..., 9:

In these example queries, the initial Group ID digit specifies a node style. You can specify that nodes in Group 2 are yellow circles, nodes in Group 4 are cyan boxes, and so on. (Having no initial ID is completely equivalent to having ID 0.)
(This is an inactive example window. To use PubGraph, please enter queries on the Basic Query page.)
Again, each of these five lines in this text area specifies a single PubMed query. However, this time the result of running with PubGraph is a page including a graph representation like this:
association graph
PubGraph allows you to use any web-standard color. However, usually it is better to use lighter or brighter colors, since they avoid affecting the readability of the node labels.

With a little experience you can produce a large variety of kinds of information about associations in the literature.

More Information

More examples of the use of PubGraph are in the Gallery.

Further information about queries is available on the PubGraph Query page and the Advanced Query page.

Becoming an expert at PubMed

To make best use of PubGraph, it helps to learn about the many capabilities of PubMed; information about this is scattered in many places, but some helpful resources include:
Two important quirks: "Quick tours" about specific kinds of queries are available at: Searching PubMed.


PubGraph has been designed to give an interesting high-level overview of the biomedical research literature. There are many possible uses -- gaining perspective on the state of a field, automatically obtaining ontology-like models of relationships between specified topics that highlight factors of interest, finding patterns in the structure, history, evolution, and impact of research fields. Achieving a good balance has required several innovations in the modeling and visualization of associations.

PubGraph is a project of the UCLA Center for Cognitive Phenomics.
Copyright © 2006-2007 D.S. Parker.