OSCAR, the chemical data checker

I spent yesterday reviewing the data in BJOC (the Beilstein Journal of Organic Chemistry) (articles). This is a new (ca. 1 year) and important journal as it is the first free-to-author and free-to-read journal in chemistry, supported by the Beilstein Institut.
BJOC advocates Openness and has a facility for adding comments on each article. Up till now there has been little or no use of this facility (it is always daunting to be the first commentator) so I thought I would add some to catalyse the process of communal knowledge generation. I’m not an expert in most of the actual science, but I am familiar with the type of data in organic papers, which has a large emphasis on describing chemical reactions and the products.
.
.
.
.
.
.
[This space blank because the images are necessarily large and overlap the wordpress menu]
.
.
.
.
.
.
There is a fairly formal manner of reporting the chemical data:

Not bedtime reading (except for chemists) but it is a formal statement of the experiment. (The tragedy is that it is used to be a machine-readable file with 16,000 points and that all chemical publishers require it to be abstracted into this text – but that is a different rant). The chemist has to extract this data from the spectrum, retype it into this form (taking ca. 45 minutes – my informal survey) and then sent to the editor. There can easily be 50 chunks of this stuff in one manuscript.
Because of this (absurb) retyping (instead of depositing the raw data with the publisher), errors creep in. The reviewer now has to check all this stuff by hand (again taking 30 minutes). And they really do – I have seen them. It could easily take a day to wade through the experimental section in a paper. Because this is the ultimate touchstone as to whether the correct compound has been made. (There have been frequent cases where claims of a synthesis have been later disproved by using this information – there was a recent example at the ACS meeting – entertainingly reviewed by Tenderbutton (search for “La Clair” in the text) where computer calculations on the proposed chemical structure did not agree with the published values). So they matter.
But what if they are mistyped? That’s where OSCAR comes in. Five years the Royal Society of Chemistry supported two undergraduate summer students (Joe Townsend and Fraser Norton) to look into creating an “Experimental Data Checker” for this sort of material. They did brilliantly and were followed next year by Sam Adams and Chris Waudby (also RSC funded). The result was OSCAR – the Open Source Chemical Analysis and Retrieval system, written in Java. Even if you are not a chemist, you may enjoy trying it out. You can take a raw manuscript (DOC) or published (HTML) – PDF is a hamburger so it probably won’t work – and drop it into OSCAR. OSCAR parses this text (using regular expressions) and produces:

OSCAR has recognised the different sort of spectra (coloured) the melting point, etc. It struggles with the appalling diversity of character sets used by Word (black squares) but makes sense of almost anything. These were very conscientious authors (they are from GSK) and the syntax is very correct. (This is impressive as there is no formal definition of the syntax, and OSCAR guesses from a large number of dialects).
OSCAR can now extract the data from the publication. Not, unfortunately, the raw data as the publishers currently don’t accept this for publication. But the numbers in the document. OSCAR can even guess what the spectrum might have looked like before it was lost in publication.

It’s only a guess, but hopefully it brings home what is lost in publication.
Now, although we don’t know whether any of the data are correct (only the author does) OSCAR has rules that point out when they might be incorrect. For example if a melting point of -300 deg C is reported it is obviously wrong (maybe a minus sign crept in). Or if the number of hydrogens calculated from the spectrum above don’t agree with the formula again something may be wrong. OSCAR applies these to each report of a chemical in the paper and lists all the warnings. Like CheckCIF (previous post) some of these are potential serious, while others are matters of style.
And, finally, OSCAR can extract all this data. I contend that the text above is factual material and cannot be copyrighted so OSCAR can extract data from the world’s chemical literature, even if it is closed and copyrighted. (However some publishers will probably claim that their articles are “entries in a database” and hence copyright. We have to fight this. It’s OSCAR that has done the work, not the publishers). The good news is (a) most data is now in open, non-copyright “Supplemental Material”, “Supporting Information”, or similar and (b) there are (a few) open access articles in chemistry. The next post shows how OSCAR can review the Open chemical literature…
(NB. This is OSCAR-1. The latest version, OSCAR3, has a much more sophisticated parser and can also work out chemical strucures from their names. It really is our vision of a “journal eating robot”. It’s on SourceForge but is strictly early adopter. More about that later.)