Yesterday I wrote about the potential advantages of using Dictionary Look-Up and Optical Structure Recognition over the “copying of molecules” from a manuscript or webpage. I noted that the copying of structures might be prone to loss of “fragments” and suggested that might have happened with discodermolide. Liquidcarbon has confirmed that was the case…the original structure came from Org. Proc. Res & Dev.
I’ve continued to look at other examples. Here’s one very complex structure…the whole structure is not visible. the images on the right are SCANNED images and those on the left are the converted connection tables. Very damned impressive I would say. If the structures are examined in detail a couple of hydroxyl groups ARE dropped as well as a couple of wedge bonds being dropped. They can be edited in easily.
The conversion process is not perfect and manual checking is necessary. But it took a few seconds to open the image and convert to a connection table. Try drawing the structure at the bottom in a few structures.
In regards to my previous post on drawing, versus look-up versus software conversion my comments have stirred a response.
My post was written with the intent that there are better ways to reproduce structures from articles that redrawing. Redrawing can introduce, as acknowledged by liquidcarbon, errors. Look-up is less prone to errors provided that work has been done ahead of time to provide a highly curated data source. I wanted to expose the scan-and convert approach because I believe it could be the process by which we convert structures from patents into connection tables for the ChemSpider database. It’s an interesting technology. The error happened to be on discodermolide…it could have been on Penicillin G or THC…the fact is that look-up or scanning are likely more efficient than redrawing and less prone to errors. BUT, both have their problems…there is no holy grail. I’ve already blogged many times on Taxol and, going with “Taxol is important” have shown that there is confusion out there (1,2,3). As I commented there…Wikipedia was linked to the incorrect structure on PubChem (but it’s edited now). The community, of which i am a member, has to have a fine eye for detail and work collaboratively to help clean up the historical records.
By the way, hexacyclinol only has five hits in Pubmed but is fairly well known in the blogosphere and is being used as the foundation for grant writing …
Now, the Agents and Eyeballs approach is very interesting to me.I like it! It’s the community approach to cleaning up the data. We’ve already provided a path to curate data online at ChemSpider. We HAVE agents running. We HAVE eyeballs reviewing. All we ask to do so is be a registered user. We would prefer non-anonymity for people when they are curating the data. It allows curators to be aware of who’s doing what. IsÂ the process optimal yet? No, I don’t think so. But it’s underway already.
This week I gave a presentation to the PubChem Working group. I published this tonight here. In this presentation I discussed the need for curation of data (slides 15-20). Slide 15 shows somebody actively curating data on ChemSpider – scroll to the bottom of this record. Slide 16 shows the results of curating taxol on wikipedia (left hand side before; right hand side after). Slide 17 is the list of synonyms for 1-butanol on an online database and originating at PubChem. Slide 19 is curation on ChemSpider…see the list of identifiers here.
I happen to think the efforts to clean up date are important. People curating data are some of my heroes. Some people talk about highly curated data. Some people help create highly curated data. Vision is great..execution is better.
I’ll recognize some of these heroes in a later blog…