Introduction
To the editor:
Anyone who regularly reads life science literature often comes across names of genes, proteins or small molecules that they would like to know more about. To make this process easier, we have developed a new, free service called Reflect (http://reflect.ws) that can be installed as a plug-in to web browsers, such as Firefox or Internet Explorer. Reflect tags gene, protein and small-molecule names in any web page, typically within a few seconds and without affecting document layout. Clicking on a tagged gene or protein name opens a popup showing a concise summary that includes synonyms, database identifiers, sequence, domains, three-dimensional structure, interaction partners, subcellular location and related literature. Clicking on a tagged small-molecule name opens a popup showing two-dimensional structure and interaction partners. The popups also allow navigation to commonly used databases. In the future, we plan to add further entity types to Reflect, including those outside the life sciences.
As science uncovers the intricate interconnections within biological systems, many life scientists constantly come across unfamiliar biochemical entities (e.g., genes, proteins or small molecules) that were previously not known to be relevant to a given field, but where today's literature shows an important, new connection. For such cases, it is clearly valuable to systematically tag all scientific entities in a publication, thus helping the reader to navigate to more specific information about any entity of interest. Such tags can help the reader to comprehend scientific content more rapidly and completely. Even when an entity is already familiar to a reader, it can be valuable to have quick access to commonly used source data entries; for example, protein sequences or two-dimensional structures of small molecules.
In spite of the clear value of systematically tagging scientific entities, only a small fraction of the main scientific publishers currently offer such tags on their web content. Some publishers are beginning to explore the option of adding tags as part of the publication process1; however, enforcing, validating and updating these tags creates additional work for publishers and authors.
The task of accurately tagging biochemical entities automatically is very challenging; this task has been the subject of intense research efforts that has led to significant improvements in accuracy2. These automated methods have been used to develop a wide variety of text mining applications and services, many of which are designed to provide sophisticated search, analysis and presentation capabilities3. However, a few text mining services have been designed to appeal to the broader life science community; for example, iHOP4 provides simple search, navigation and presentation of Medline abstracts with systematically tagged gene and protein names.
Tagging a scientific entity is only half the story: the other half is the information that is accessed when the user clicks on a tag. In the past, entity tags were almost always simple hyperlinks to web pages showing source data entries. Increasingly, however, entity tags are not hyperlinks but scripts that create a small popup window (typically with Javascript). A key advantage of using popups is that users can see basic information about an entity without having to navigate away from the current web page. If needed, hyperlinks to more detailed information can be provided on the popup.
An emerging trend is to augment normal web browsing by using plug-ins, such as Greasemonkey (http://greasespot.net/), that let end-users modify the appearance of web pages while browsing. We believe that such augmented browsing tools will soon have an important impact on how scientists read literature on the web. For example, one such tool, ChemGM5, lets end-users tag small-molecule names in any web page; clicking on a tagged small molecule opens a popup that shows the two-dimensional structure. Tagging is done by sending the page to a remote server, and the total time taken is typically about one minute for a five-page document. Another tool, Concept Web Linker (http://conceptweblinker.wikiprofessional.org/), has a broader scope: it tags a range of entities, such as genes, chemicals and diseases, again typically within about one minute. However, the Concept Web Linker popups show less specific information, giving only a short text description for each entity; to reach more specific information, such as protein sequences, the user needs to navigate through a series of web pages, in some cases browsing complex ontologies. A related system, Cohse6, has even broader scope—it enables users to choose many different ontologies, including those outside the life sciences. Currently, however, the publicly accessible versions of Cohse provide only very limited functionality and using the life-science ontologies provided does not allow direct navigation to specific information, such as sequences.
We designed Reflect to be an augmented browsing tool that would be broadly useful to life scientists, and would address the limitations of the above tools. A primary goal of Reflect was to enable the user to navigate directly from a gene or protein name to a specific sequence. A second goal was to be able to tag a typical web page in a few seconds. A third goal was to provide entity popups that give a concise summary of the most important features of the entities, as well as direct hyperlinks to commonly used source data entries (Fig. 1 and Supplementary Methods online). Finally, Reflect was designed with a strong focus on ease of installation and on usability.
Figure 1: The Reflect button can be installed in the Firefox or Internet Explorer web browsers.
Clicking the Reflect button tags protein and gene names (blue highlighting), and small molecules (orange highlighting) in any web page. Clicking on a highlighted name opens a small popup showing a concise summary of important features of the entity, and provides access to related information (Supplementary Methods).
Full size image (126 KB)Reflect can be used directly from http://reflect.ws/ by typing or pasting in a URL. In this case, the Reflect server retrieves the HTML document, tags it and returns the tagged version to the user's browser. Note that this will work only for URLs that are publicly accessible.
A more convenient way to use Reflect is to install it as a plug-in to Firefox or Internet Explorer. In this case, the HTML document is retrieved by the user's browser, then sent to the Reflect server, tagged and returned to the browser. Thus, with the plug-in, users can 'Reflect' any page that they can access.
The Reflect server at the European Molecular Biology Laboratory keeps in RAM (random-access memory) a large dictionary with names and synonyms for 4.3 million small molecules, and for 1.5 million proteins from 373 organisms. When tagging an HTML document, the server finds all occurrences of these synonyms and returns a slightly modified version of the HTML document to the user's browser—the only difference is that all matching protein, gene and small-molecule names are now tagged and highlighted. Tagging a document usually takes much less time than uploading and downloading it; thus, the time taken for the entire process (upload, tag and download) depends almost exclusively on the speed of the user's internet connection. With standard broadband, the entire process usually takes from one to five seconds for a five-page document (Supplementary Methods).
Clicking on a tagged small-molecule name opens a summary popup (Fig. 1, bottom right) that shows two-dimensional structures from PubChem7 and interaction partners from STITCH8. Clicking on a tagged protein or gene name opens a popup (Fig. 1, top right) that shows synonyms, the complete amino acid sequence of the longest transcript, domains from the SMART9 database, a representative three-dimensional structure from PDBsum10, principal interaction partners from STITCH8, known subcellular location and an image of the organism. Most of these features on the popup are hyperlinked to related database entries. The popup also has hyperlinks to the corresponding gene entry and to related Medline abstracts in iHOP4. Dragging the mouse on the domain graphical view scrolls through the sequence, and hovering over a domain causes the domain name to appear in a tool tip.
When a tagged name is ambiguous, the popup shows all possible matches and allows the user to disambiguate the name by choosing which of the possibilities is most appropriate. Currently, three levels of ambiguity are shown. First, a name may match both a protein and a small molecule; in this case, Reflect shows both possibilities on separate tabs. Second, a name may match to several genes within the same organism; here, Reflect shows all matching genes in a pull-down menu. And third, for gene and protein names, it is often ambiguous which organism is intended in the HTML document; to address this, Reflect shows a list of possible organisms derived from the default organism (which is initially set to human, but can be changed using the Firefox plug-in) plus organisms mentioned in the document. In the near future, we also plan to show a fourth level of ambiguity, where users will be able to select splice variants for each gene.
Any automated method for recognizing biochemical entity names will make some errors: some false positive matches will arise due to overlap with commonly used words or acronyms, and false negatives will arise due to incompleteness of the tagging dictionary. To assess the accuracy of Reflect, we tested it against the BioCreative11 benchmarks. Compared with 15 other tools for automated entity recognition that were assessed in BioCreative, Reflect ranked second best (91% F-score) using the Saccharomyces cerevisiae benchmark and had median performance (66% F-score) using the Drosophila melanogaster benchmark. We consider these to be quite good results because, unlike the other tools tested against these benchmarks, Reflect was designed to optimize speed rather than accuracy.
In the near future, we plan to enable community-based, collaborative editing for some of the information in Reflect popup, especially the synonym lists. These and other planned extensions will enable the user community to improve Reflect by correcting false-negative and false-positive matches. We plan to add further entity types (e.g., diseases, pathways and organisms), and eventually to add entity types beyond the life sciences; we designed Reflect to be an extendible platform, and we welcome collaboration proposals for adding further entity types. In addition, we welcome proposals from publishers and data providers interested in programmatic access to Reflect. With such access, end-users can use 'Reflected' content without needing to install a browser plug-in.
ADVERTISEMENT
In summary, Reflect creates a view of the web tailored for the life scientist, that is, with systematic tagging of biochemical entities, and easy access to more detailed information. Reflect is already being used by thousands of researchers, and we have received much positive feedback regarding Reflect's usefulness and ease of use. In addition, just before publication of this correspondence, Reflect was awarded first prize in the Elsevier Grand Challenge, a contest for tools that improve the way scientific information is communicated. Thus, we believe that Reflect can be a valuable tool for researchers, teachers, students and anyone who reads life science literature on the web. We further predict that in the near future tools such as Reflect will change dramatically how scientists use the web.
Note: Supplementary information is available on the Nature Biotechnology website.
