I hope that the question is self-explaining.
Egon Willighagen
[ Admin ]
You can also look a this from another perspective. Go for the quantity, and do the validation yourself. Actually, you would be wise to do this for the smaller, curated databases too. Your particular use case is rarely an exact match with the database; therefore, you are likely in the situation where you have to preprocess your input anyway. Why not just include the validation in that process?
This validation is pretty easy to do with one of the many Open Source utilities around. For example, the CDK can validate many aspects, and do other kind of filtering. Diversity analysis will not be a problem with the free tools around either (e.g. What tools are available for clustering
small molecules ?).
I partially agree. But… How can you validate with open source tools (or commercial tools) if the level of ambiguity is too high at the origin? Look at these structure: http://alchemoinformatics.blogspot.com/2010/05/which-is-real-ru-486-2.html. Is this a major issue or not?
@matteo floris This paper just came out: “Trust, But Verify: On the Importance of Chemical Structure Curation in Cheminformatics and QSAR Modeling Research” –> http://dx.doi.org/10.1021/ci100176x
Gilbert Müller
[ Editor ]
As so often, the “best compromise” depends on time and effort one can
spend in a project, (as you most probably know (:–) ).
However, some of my starting points come from ZINC :
NCI Diversity Set II (1364 compounds, also availably here)
NCI Plated 2007 with cluster representatives at 60% (6400 compounds)
PubChem catalog with cluster representatives at 60% (~ 20k compounds)
Commercial, but IMHO still interesting is the emolecules database.
Regarding quality there are many 3D structures out there,
but usually I do these calculation on my own to be sure
that needle and haystack structures are treated the same way…
The interesting thing about the eMolecules database is that they don’t include any chemicals which are not immediately available for purchase. Vendors that are slow to ship, or don’t have the chemicals in stock, are removed after a warning.
@baoilleach unless things have changed eMolecules had the NIST Webbook data and some of those molecules are not available from any vendor…like the sodium chloride dimer that PMR took me to task on: http://www.chemspider.com/blog/?p=6
Imants Zudans
ZINC is good choice. If you are interested in unprocessed structures or lightly processed structures then I think eMolecules offer download of all of their data of commercially available compounds. My company MolPort also works with computational chemists: see our press release with InhibOx. We provide for free all our compound database in SD file format. Once you finish the virtual screening and are ready to test a set of compounds, we can source them for you for supplier direct prices.
Antony Williams, ChemConnector
[ Editor ]
I guess the question is what is the measure of “quality”? Is it is accuracy of the data associated with the chemical structures, the links to compounds which are still commercially available or the structure depictions? There are many interpretations of quality in this space. In terms of validated data associated with chemical compounds I believe that ChemSpider (www.chemspider.com) is the largest source of data with validated content that is online. It is not all validated yet as it is over 24.7 million compounds. Most people use ZINC which I think is likely the most useful for your work as it includes a large number of commercial collections. eMolecules does contain a lot of commercially available compounds but also has the NIST webbook collection that contains a lot of obscure chemistry such as the sodium chloride dimer (see here: http://www.chemspider.com/blog/aggregated-chemistry-and-quality-is-chemspider-a-good-representative.html)