Thursday, July 10, 2008

A Gene Wiki

The prevalence of errors and omissions in sequence databases is one of the ugly little secrets of molecular biology [Errors in Sequence Databases]. We know how to fix the problem; it requires careful annotation by knowledgeable experts. Unfortunately, this is time-consuming and expensive since you have to hire annotators. One other possibility is to allow open access to all existing records in databases such as GenBank, RefSeq, or PDB. This ain't gonna happen because here's no way to verify the changes to make sure they are valid. The people who control these databases are very reluctant to allow open access and the authors of the database entries are uneasy about allowing others to insert annotations into their records.

But there are other models that might work. A recent paper by Huss et al. (2008) in PLoS Biology describes a possible solution. They point out that Wikipedia seems to be a successful model of collaborative effort to ensure accuracy. Why not adopt this model for gene annotation?

Some examples of human genes already had Wikipedia entries and these entries were updated and annotated by various users. In order to stimulate and encourage this process, Huss et al. (2008) created stub entries on Wikipedia for every human gene. Here's how they describe it in their paper.
In principle, a comprehensive gene wiki could have naturally evolved out of the existing Wikipedia framework, and as described above, the beginnings of this process were already underway. However, we hypothesized that growth could be greatly accelerated by systematic creation of gene page stubs, each of which would contain a basal level of gene annotation harvested from authoritative sources. Here we describe an effort to automatically create such a foundation for a comprehensive gene wiki. Moreover, we demonstrate that this effort has begun the positive-feedback loop between readers, contributors, and page utility, which will promote its long-term success.
Today, anyone with access to Wikipedia can contribute to annotating human genes. Two examples of well annotated genes are HSP90 and NF-κB.

Let's look at some examples of stub entries to see how the process might work. I've chosen the human members of the HSP70 multigene family because I'm familiar with these genes. All members of the family function as molecular chaperones, helping to ensure that proteins are properly folded [Heat Shock and Molecular Chaperones].

There are two major inducible genes called HSPA1A and HSPA1B. They are adjacent to one another on chromosome 6. The database entries for these genes are confusing and in most cases it's almost impossible to discern which gene is being referred to.

Here's the Wikipedia stub for HSPA1A. Clearly there's an opportunity to modify this entry in order to make it clear that there are two very similar genes and to point to the proper sequence records for this gene. The second gene, HSPA1B, has its own entry in EntrezGene so I was expecting to find it on Wikipedia. Unfortunately, it's not there. A search for HSPA1B redirects you to HSPA1A. So right away we have a problem. Someone made a decision to merge these entries on Wikipedia making it very difficult to correctly annotate the separate genes.

HSPA1L is an intronless gene closely linked to HSPA1A and HSPA1B. HSPA1L is not heat shock inducible, instead it is developmentally regulated. The gene is expressed exclusively in the testes. The stub entry for this gene [HSPA1L] includes an RNA expression profile that beautifully illustrates the developmental regulation but there's nothing in the annotations that mentions this. This is an excellent opportunity to correct an omission in the existing databases.

Let's look at one more example to see how useful the Wikipedia effort might be. The HSPA4 gene is identified on all databases as a member of the HSP70 gene family. It's usually called "Heat shock 70kDa protein 4." The Wikipedia stub reflects the GenBank annotation [HSPA4]. However, it has been known for a long time that this gene is NOT a member of the HSP70 gene family. The annotation is incorrect. Instead, this gene is Apg-2 an HSP100 homologue not related to HSP70. The original error is due to Fathallah et al. (1993) who sequenced the first example. They mistakenly called it a novel hsp70 gene due, in part, to sequencing errors and partly to an overactive imagination. Mistakes such as these are extremely difficult to remove from the database but we now have an opportunity to correct the error on the Wikipedia entry.

Putting the human genes on Wikipedia is almost as good as allowing open access to the primary sequence databases. The effort will only be successful if scientists make the effort to edit the Wikipedia entries. It's unlikely that most gene entries will be modified but even if only a subset is annotated, it's better than none at all. It would be nice if the RefSeq records could point to the Wikipedia records. That will encourage people to make comments on Wikipedia.

Huss III, J.W., Orozco, C., Goodale, J., Wu, C., Batalov, S., Vickers, T.J., Valafar, F., and Su, A.I. (2008) A Gene Wiki for Community Annotation of Gene Function. PLoS Biol 6(7): e175 [doi:10.1371/journal.pbio.0060175]


  1. Wikipedia is an opinion poll, not a factual reference.

  2. I disagree with the above comment. Sure, for contentious and current events, you find all sorts of opinions creeping up into the wiki entries. But generally, most the core science articles are pretty good and even insightful sometimes. That's probably because opinions tend to get weeded out by scientific facts eventually. However, I still can't say that I'd trust wikipedia over a peer-reviewed published source. It has reached that level of reliability, and I don't know if it can. Certainly, more involvement by scientists to police the entries they are experts in would be helpful. Issues like these have been discussed in Science and Nature magazine. Perhaps what is needed is some kind of method of assigning particular wikipedia entries to known experts on the topic.

  3. Thanks for having a look at the Gene Wiki and your thoughtful post.

    I did a little digging and I still can't figure out why HSPA1A and HSPA1B do not have distinct pages. Clearly from the Entrez Gene pages, both symbols are synonyms of the other, underscoring the confused nature that you allude to. But our program should use Entrez Gene IDs, and as you note, those are different. We are looking into this one. Hopefully we'll get a page for HSPA1B soon.

    Agreed that links *to* the Gene Wiki pages will greatly accelerate their growth. One problem to accomplishing this is that the gene pages in Wikipedia do not appear at systematic page titles, complicating the linking from other databases. Although we neglected to mention this in the paper, we do have a service that will take an Entrez Gene ID and route you to the appropriate Wikipedia page. For example, I can find the correct page for the gene ITK (Entrez Gene ID = 3702) by using this URL: Similarly, finding the page for HSPA1A (3303) can be found using

    Finally, regarding your comments on HSPA1L and HSPA4, there's only one last step that would take this thought experiment and turn it into an actual contribution to the Gene Wiki. Have you thought of contributing your insights and expertise to those pages?

  4. @J, the Gene Wiki will definitely never replace peer-reviewed sources. We think of this as a complementary tool. Clearly, data is not evaluated by an expert (e.g., a gene portal curator, or a peer reviewer), but that also means that a curator or peer reviewer is not on the critical path to adding new content. If you see something incomplete/misleading/incorrect, you have full power to fix it.

    We hope that scientists will understand that the open model of Wikipedia requires a different mindset with reading the Gene Wiki articles. It's also worth noting that Wikipedia allows and encourages inline citations. Check out, for example. Many statements there are footnoted, and those footnotes often link directly to Pubmed. These links obviously can be very helpful when readers try to evaluate the reliability of any given statement.

  5. It's a good idea and I will make sure to contribute for a few proteins I know well.

    My gripe is with the ribbon models shown. They need different labels to indicate that the PDB is that of a fragment of the whole protein!

    Right now it's ridiculous. Take plectin, a 500 kDa protein. The pic shows a small fragment (PDB code 1mb8), which is only ~ 25 kDa. Yet, the title says "Plectin 1, intermediate filament binding protein". Funny for me but probably misleading for most.

    Also, Wikipedia editing has to get a WYSIWYG interface. I don't have to go through codes and tags when using Word - why do I have to to edit Wiki pages? I am not about to learn HTML just to be able to figure where the heck is that figure title is located...

  6. @DK, Good point on the plectin structure. Unfortunately, there's only room to show one structure (of the five that are available, shown hyperlinked under the image), and right now the bot just chooses the first one alphabetically. Among many other things, this is clearly an area that could be improved.

    Nevertheless, this type of enhancement arguably speaking is exactly the type of edits we hope will come from the community at large. After all, what the "best" representative structure is probably best defined by a human, not a computer. Looking at the five options, is 1SH5 the one you would choose? If so, I will make the change for you (since edits to the infobox are a bit more difficult than editing the main text area) and then post step-by-step instructions for you or others to change other gene pages.

    I hear you on the WYSIWYG editor. It's a bit surprising to me that more hasn't been made of this. But, for the Gene Wiki, we're stuck with what Wikipedia uses. If you're editing text, I find using your browser's search function helps to get me to exactly the place I want to start editing and bypass unrelated gibberish. Hope that helps.

    More feedback is welcome and appreciated, either here or at

  7. Out of little more than idle curiosity, I looked for the gene NKX3-2, and didn't find it. I'm not a scientist, so I may be mistaken about this, but I think that it is a real human gene, and it does seem to be on other databases. (And I wouldn't be so bold as to make an entry for it!) As I say, just out of curiosity.

  8. @Anonymous, you are right, a page doesn't exist for NKX3-2. We limited page creation to the top ~9000 genes (when sorted by the number of citations in the literature) to satisfy Wikipedia's notability criteria. NKX3-2 came in just outside our threshold, tied at #9875.

    We're planning to create a web site where people can submit any gene id to create a page for their gene of interest, regardless of where it ranked on our list. But that web site is still in progress...

  9. To: Andrew Su

    Andrew, just IMHO: whenever the structure represent only a *fragment* of the full protein, the title should state this explicitely. Ideally, with a the domain name. Frequently different domains of the same protein are available seprately. In which case there can be no "best" and both should be shown.

    Speaking of plectin only as an example, here is what's available:

    DBREF 1MB8 A 59 293 UNP Q15149 PLEC1_HUMAN 59 293

    DBREF 1SH5 A 7 243 UNP Q9QXS1 PLEC1_MOUSE 181 417

    1SH6 is the same as 1SH5

    DBREF 2ODU A 300 530 UNP Q6S383 Q6S383_HUMAN 300 530

    2ODV is the same as 2ODU

    In other words, from human protein we have S10-plectin domain + first CH domain (1MB8) and first SPEC domain (2ODU). From mouse we have two CH domains in tandem (1SH5). Loking at the structures, it is quite obvious that the s10-plectin domain identified by sequence homology has a fold that is identical to CH domains (and thus should really be called CH domain).

    So the fullest picture of the N-terminus of plectin that we have now looks like three tightly connected balls (CH domains) followed by a rod (specrtin fold) connected to the balls. This accounts for about 1/3 of the protein. Another 1/4 of the protein is obviously longer rod (more SPECs) that ends with the rest of the protein that is a bunch of PLEC "domains" with yet unknown structure.

    Off-topic speculation: plectin is not a dimer.
    1) alpha-beta spectrin antiparallel dimer makes no sense if you want to join thin and intermediate filaments; 2) if it were a parallel dimer, it would have had huge affinity for F-actin (cooperativity between 4-6 CH domains) - that's not the case; 3) its domain structure resembles dystrophin (CHs + SPECs), which was thought to be a dimer based on "paper biochemistry" but was experimentally proven to be a monomer physiologically.

    Not to be picky but besides OMIM I would like to also see links to the respective Uniprot, HPRD and SMART entries. This would make exploring much easier. where available, SNPedia links would be great too (

  10. @andrew su:
    "We're planning to create a web site where people can submit any gene id to create a page for their gene of interest, regardless of where it ranked on our list. But that web site is still in progress..."

    If I understand you correctly, what you are saying is that the plan would be that there would be a web site where I could enter NKX3-2 (for example), and then it would generate a Wikipedia entry from the existing information, which would have details somewhat comparable to any of the present entries?

  11. @DK, thanks for your comments. You raise many valid points about how we deal with protein structure. I'm sure version 2.0 of our effort will give us ample opportunity to deal with these and other issues. ;)

    On to a couple of specifics. You echo some previous comments from Wikipedia members that we should show protein domain information in the infobox. The challenge is to include this information (and the other links to external databases) without overwhelming the user with information. In some cases, we'll have to be content with having these resources be two clicks away from the wikipedia page. (Of course, if they are particularly relevant to a specific gene, users can add links at the bottom of the free-text section under "External links".)

    Second, I see the confusion around the infobox title. We actually mean for that title to refer to the official gene/protein name, not to the PDB image. For example, consider TNFRSF1B. Even though it has no protein structure, the infobox has the full name there. In any case, we'll think about ways to make this more clear (and suggestions are welcome).

    Lastly, anyone is welcome to participate on the bot's discussion page at, which is where we do most of our brainstorming and discussions.

  12. @Anonymous, yes, you have interpreted it exactly right, with only one minor difference. We are using a previous tool as a model. See as an example. In this tool, you can enter a gene ID (which I've already filled in), and then the web page outputs formatted "wikicode". The user can then simply copy-and-paste into a new or existing article, and that will give it the same basic layout as all the other gene pages we've automatically created. Hope that makes sense, and suggestions are of course welcome.

  13. @andrew su:

    OK, I just tried it, and produced an entry in the "Sandbox". I'll leave it at that, and not try to get in the way of the professionals.


  14. A splendid idea (to separate interpretation from data [bases]).

    And I would like to thank Andrew Su for a promising explanation for how this can be made to work.

  15. Have you considered Freebase? Freebase is a freely editable database, which has imported data from Wikipedia, and allows users to define their own types. See e.g. their definition of Gene:

  16. @luke, yes, freebase is also an interesting tool that I think biologists should consider. I think freebase and the Gene Wiki are complementary. Freebase is primarily focused on harvesting structured content from its users, wherein each contributed piece of data is "typed". This property aids efforts on downstream mining, but it does require some additional effort on the part of the contributing biologist, and some biologists won't take that time. In contrast, the Gene Wiki focuses on unstructured content (free text, images, diagrams, photos, etc.). This format is more difficult for a computer to mine, but the output is more "human readable" (since it is text like you would read in an article) and it makes contributing a bit easier for the biologist. Again, they are complementary models, and I think biology as a field can use many different approaches for harnessing community intelligence.

    @Torbjörn, thanks for your kind comments. Now, have you edited the page on your favorite gene yet? ;)

  17. I see that someone has added a page on my favorite gene, NKX3-2. Thanks to whoever did that.

    My interest is that there seems to be a connection between NKX3-2 and the evolutionary change from jaw to middle ear.

  18. @anonymous, it is the wikipedia way... Someone probably saw your comment in this blog and just did it, invoking the "Be Bold" policy at Wikipedia. Now that it's done, consider adding a comment about it. (Though keep in mind that all contributions should be backed up by verifiable sources and should not reflect personal hypotheses.)

    (And actually the editor who created that page, Boghog2, is clearly one of Wikipedia's most tireless contributors, so it may be that he created that page entirely through his recent and extensive work on transcription factors. Click the "show" link at the bottom of the NKX3-2 page to see all the pages which have been categorized as transcription factors in Wikipedia.)

  19. You're on the Molecular and Cell Biology Carnival -- here:

    Thank you for submitting your article!