Showing posts with label Data publishing. Show all posts
Showing posts with label Data publishing. Show all posts

Thursday, 17 December 2009

More activity on semantic publishing

If you saw tweets from @cardcc today, you might realise I’ve been very interested in a couple of recent developments in semantic publishing. I wrote earlier about linking data to journal articles, including David Shotton’s adventures in semantic publishing. David’s work was one of those included in the review article in the Biochemical Journal by Attwod, Kell, McDermott et al (2009). The article ranged over the place of ontologies and databases, science blogs, and various experiments. These included

  • RSC and Project Prospect,
  • The ChemSpider Journal of Chemistry,
  • The FEBS Letters experiment,
  • PubMed Central and BioLit,
  • Shotton’s PLoS experiment,
  • Elsevier Grand Challenge,
  • Liquid Publications,
  • The semantic Biochemical Journal experiment.

The latter was the real focus of the article, available in PDF, but which if read through a special reader called Utopia Document displayed some active capabilities. These included the ability to visualise and rotate 3-d images of proteins, to see tables represented as graphs (or vice versa) and to link to entries in nucleic acid databases. The capabilities were perhaps a bit awkward to spot and to manipulate, but still interesting. This article is (gold) open access. Other articles in the issue have also been instrumented in this way.

It’s clearly early days for Utopia, and I wasn’t wholly impressed with it as a PDF reader, but I was certainly very excited at some of what I read and saw.

I also read today a very different article (I think not available on open access), by Ruthensteiner and Hess (2008). They describe the processes in making 3-d models of biological specimens, and presenting them in PDF, readable by a standard Acrobat Reader. The 3-d capability was at least as good as if not better than the Utopia results.

Because it’s getting late, I’ll end with my last tweet:

“My head is spinning with semantic article possibilities. I hope some get picked up in new #jiscmrd proposals, see http://www.jisc.ac.uk/fundingopportunities/funding_calls/2009/12/1409researchdata.aspx"


Attwood, T. K., Kell, D. B., McDermott, P., Marsh, J., Pettifer, S. R., Thorne, D., et al. (2009). Calling International Rescue: knowledge lost in literature and data landslide! The Biochemical journal, 424(3), 317-33. doi: 10.1042/BJ20091474.

Ruthensteiner, B., & Hess, M. (2008). Embedding 3D models of biological specimens in PDF publications. Microscopy research and technique, 71(11), 778-86. doi: 10.1002/jemt.20618. Pubmed abstract.

Wednesday, 18 November 2009

Data and the journal article

I recently had a discussion (billed as a presentation, but it was on such an (ahem) intimate scale that it became a discussion) at Ithaka, the organisation in New York that runs JSTOR, ArtSTOR and Portico. We talked about some of the issues surrounding supporting journal articles better with data. Both research funders and some journals are starting to require researchers/authors to keep and to make available the data that supports the conclusions in their articles. How can they best do this?

It seems to me that there are 4 ways of associating data with an article. The first is through the time-honoured (but not very satisfactory) Supplementary Materials, the second is through citations and references to external data, the third is through databases that are in some way integrated with the article, and the fourth is through data encoded within the article text.

My expectation was that most supplementary materials that included data would actually be in Excel spreadsheets, and a few would be in CSV files, while even fewer would be in domain-specific, science-related encodings. I was quite shocked after a little research to find, at least for the Nature journals I looked at, that nearly all supplementary data were in PDF files, while a few were in Word tables. I don't think I found any that were Excel, let alone CSV. This doesn't do much for data re-usability! As things stand currently, data in a PDF document (eg in tables) will probably need to be extracted by hand copy; possibly by cut and paste followed by extensive hand manipulation.

I would expect that looking away from the generalist journals towards domain-specific titles, would reveal more appropriate formats. However, a ridiculously quick check of Chem-Comm, a Royal Society of Chemistry title, showed supplementary data in PDF even for an "electronically enhanced article (eg Experimental procedures, spectra and characterization data, perhaps not openly accessible...).

There’s a bit of concern in some quarters about journals managing data, particularly that data would disappear behind the pay wall, limiting opportunities for re-use.

What would be ideal? I guess data that are encoded in domain-specific, standardised formats (perhaps supported by ontologies, well-known schemas, and/or open software applications) would be pretty useful. I’ve also got a vague sense of unease about the lack of any standardised approach to describing context, experimental conditions, instrument calibrations, or other critical metadata needed to interpret the data properly. This is a tough area, as we want to reduce the disincentives to deposit as well as increase the chances of successful re-use.

Clearly there are many cases where the data are not appropriate for inclusion as supplementary materials, and should be available by external reference. Such would be the case for genomics data, for example, which must have been deposited in an appropriate database (the journal should demand deposit/accession data before publication).

External data will be fine as long as they are on an accessible (not necessarily open) and reasonably permanent database, data centre or repository somewhere. I do worry that many external datasets will be held on personal web sites. Yes, these can be web-accessible, and Google-indexed, but researchers move, researchers die, and departments reorganise their web presence, which means those links will fail, and the data will disappear (see the nice Book of Trogool article "... and then what?").

Sometimes such external data can be simply linked, eg a parenthetical or foot-noted web link, but I would certainly like to encourage increasing use of proper citations for data. Citations are the currency of academia, and the sooner they accrue for good data, the sooner researchers will start to regard their re-usable data as valuable parts of their output! It’s interesting to see the launch of the DataCite initiative coming up soon in London.

There is this interesting idea of the overlay data journal, which rather turns my last paragraph on its head; the data are the focus and the articles describe the data. Nucleic Acids Research Database Issue articles would be prime examples here in existing practice, although they tend to describe the dataset as a persistent context, rather than as the focus for some discovery. The OJIMS project described a proposed overlay journal in Meteorology; they produced a sample issue and a business analysis, but I’m not sure what happened then.

The best (and possibly only) example I know of the database-as-integral-part-of-article approach is Internet Archaeology, set up in 1996 (by the eLib programme!) as an exemplar for true internet-enabled publishing. 13 years later it's still going strong, but has rarely been emulated. Maybe what it provides does not give real advantages? Maybe it's too risky? Maybe it’s too hard to create such articles? Maybe scholarly publishing is just too blindly conservative? I don't know, but it would be good to explore in new areas.

Peter Murray-Rust has argued eloquently at the tragedy of data trapped and rendered useless in the text, tables and figures of articles. We would like to see articles semantically enriched so that these data can be extracted and processed. Encoded data points us to a few examples, such as the Shotton enhanced article described in Shotton et al 2009, and also to the Murray-Rust/Sefton TheoREM-ICE approach (although that was designed for theses, I think). I think the key here is the lack of authoring tools. It is still rather difficult to actually do this stuff, eg to write an article that contains meaningful semantic content. The Shotton target article was marked up by hand, with support from one of the authors of the W3C SKOS standard, ie an expert! The chemists having been working on tools for their community, both the ICE example, and also MS Chem4Word, maybe ChemMantis, etc.

This last paragraph also points us towards the thesis area; I think this is one that Librarians really ought to be interested in tackling. What is the acceptable modern equivalent to the old (but never really acceptable) practice of tucking disks into a pocket inside the back cover of a thesis? Many universities are now accepting theses in digital form; we need some good practice in how to deal with their associated data.

So, we seem to be quite a way from universal good practice in associating data with our research articles.

Shotton, D., Portwin, K., Klyne, G., & Miles, A. (2009). Adventures in semantic publishing: exemplar semantic enhancements of a research article. PLoS Computational Biology, 5(4). doi: 10.1371/journal.pcbi.1000361.

Thursday, 10 September 2009

Nature special section on data sharing

This week's issue of Nature includes a very interesting special section on data sharing. There is a (not entirely accurate) editorial, a feature (on why data archives are so empty), and two opinion pieces.

The first opinion piece is on pre-publication deposit, linked to a workshop held in Toronto (and continuing the famous Bermuda and Fort Lauderdale workshops on data sharing in genomics) where attendees "[recommended] extending the practice to other biological data sets". The workshop produced the "Toronto Statement", which has recommendations for funding agencies, data producers, data anlysts and users, and scientific journal editors. The statement encourages rapid pre-publication deposit where the datasets are:
  • "Large scale (requiring significant resources over time)
  • Broad utility
  • Creating reference data sets
  • Associated with community buy-in"
The second opinion piece is on post-publication deposit, linked to a meeting of mouse researchers in Rome "[proposing] ways to promote a culture of sharing". It's partly about other kinds of research materials (mice and cell-lines), but mostly about data. A couple of paragraphs are worth quoting (I hope Nature doesn't mind!):
"Where they don't yet exist, clear criteria should be developed for reviewers of grants to help them assess data and material-sharing plans submitted as part of a funding proposal. There are already examples of good practice in this regard from the NIH, the Howard Hughes Medical Institute, and several UK funding organizations such as the Wellcome Trust and the Medical Research Council. Data-sharing plans are required in proposals, efforts are made to facilitate sharing, such as putting investigators in touch with repositories and, for some organizations, compliance is an important consideration in funding renewal.

"Deposition of data and resources into public repositories is important for the validation of published results, as well as facilitating reuse. Although it is usual practice for major public databases to make data freely available to access and use, any restrictions on use should be strongly resisted and we endorse explicit encouragement of open sharing, for example under the newly available CC0 public domain waiver of Creative Commons."
The latter point is particularly interesting; in the UK restrictions on sharing often do seem to exist, even if only required registration to access. Sometimes the argument here is about protecting confidentiality; sometimes it is about ensuring the archives know their customers, for sustainability reasons. So this paragraph is also interesting:
"Many of the major public data repositories have no stable underlying funding and there are data types, particularly new ones, without appropriate public data repositories. We encourage further investment and recommend that public database coverage and stability be looked at in a coordinated way by funding organizations and the community with increased urgency. A good model is provided by the UK Biotechnology and Biological Sciences Research Council's Bioinformatics and Biological Resources Fund, which provides dedicated funding for development and sustainability of public resources and informatics tools."
This all appears to be Open Access, and is well worth reading.

Tuesday, 20 January 2009

Load testing repositories

One of the issues that has worried me about moving from repositories of e-prints to repositories of data is the increased challenges of scale. Scale could be vastly different for data repositories in several dimensions, including
  • rate of deposit
  • numbers of objects
  • size of objects
  • rate of access
  • rate of change to existing objects...
Now Stuart Lewis is reporting on the first stage of the JISC-funded ROAD project, where they have load-tested a DSpace implementation (on a fairly chunky configuration), loading 300,000 digital objects of 9 MB each.

Stuart reports
  • "As expected, the more items that were in the repository, the longer an average deposit took to complete.
  • On average deposits into an empty repository took about one and a half seconds
  • On average deposits into a repository with three hundred thousand items took about seven seconds
  • If this linear looking relationship between number of deposits and speed of deposit were to continue at the same rate, an average deposit into a repository containing one million items would take about 19 to 20 seconds.
  • Extrapolate this to work out throughput per day, and that is about 10MB deposited every 20 seconds, 30MB per minute, or 43GB of data per day.
  • The ROAD project proposal suggested we wanted to deposit about 2Gb of data per day, which is therefore easily possible.
  • If we extrapolate this further, then DSpace could theoretically hold 4 to 5 million items, and still accept 2B of data per day deposited via SWORD."
They plan to repeat these tests with ePrints and FEDORA platforms.

Like all such, it's an artificial test, but it does give encouragement that DSpace could scale to handle a data repository for some tasks. I don't know if other issues would be show-stoppers or not, for something like a lab repository, but most of the scale issues seem OK.

Thursday, 6 November 2008

Data publishing and the fully-supported paper

Cameron Neylon’s Science in the Open blog is always good value. He’s been posting installments of a paper on aspects of open science, and there’s lots of good stuff there. Of course, Cameron’s focus is indeed on open science rather than data, but data form a large part of that vision. In part 3:
“Making data available faces similar challenges but here they are more profound. At least when publishing in an open access journal it can be counted as a paper. Because there is no culture of citing primary data, but rather of citing the papers they are reported in, there is no reward for making data available. If careers are measured in papers published then making data available does not contribute to career development. Data availability to date has generally been driven by strong community norms, usually backed up by journal submission requirements. Again this links data publication to paper publication without necessarily encouraging the release of data that is not explicitly linked to a peer reviewed paper.”
“In other fields where data is more heterogeneous and particular where competition to publish is fierce, the idea of data availability raises many fears. The primary one is of being ’scooped’ or data theft where others publish a paper before the data collector has had the ability to fully analyse the data.”
In part 4, Cameron discusses the idea of the “fully supported paper”.
“This is a concept that is simple on the surface but very complex to implement in practice. In essence it is the idea that the claims made in a peer reviewed paper in the conventional literature should be fully supported by a publically accessible record of all the background data, methodology, and data analysis procedures that contribute to those claims.” He suggests that “the principle of availability of background data has been accepted by a broad range of publishers. It is therefore reasonable to consider the possibility of making the public posting of data as a requirement for submission.”
Cameron does remind us
“However the data itself, except in specific cases, is not enough to be useful to other researchers. The detail of how that data was collected and how it was processed are critical for making a proper analysis of whether the claims made in a paper to be properly judged.”
This is the context, syntax and semantics idea in another guise. Of course, Cameron wants to go further, and deal with “the problem of unpublished or unsuccessful studies that may never find a home in a traditional peer reviewed paper.” That’s part of the benefit of the open science idea.