With release 1.4.8 there were some changes in the licensing policy. The code is now under GNU Affero General Public License (it was GPL so far). The data are under an extended version of the ODC Open Database License. This allows use of the data, but requires software built using the data to be under an open source license (plus the usual conditions of the ODbL). We think this is a fair policy keeping the data available, but enforcing users to contribute. The license covers all data wether published in the predictor jars, the database dump or the text file formats .
I compile statistics about usage of the web interface from time to time and I thought it might be nice to share these. So here are a few statistics and comments. Just to keep users assured that privacy is respected: No queries, personal data or IPs are published and no information which can be put together to build profiles are made public. This is just aggregate statistics we use here. Logging of queries is in line with the nmrshiftdb2 privacy policy outlined at http://nmrshiftdb.nmr.uni-koeln.de/portal/js_pane/P-Help?URL=using.html#register... read more
The standalone prediction code has been updated. It now includes the stereo prediction. This should work as describe in http://www.nmrshiftdb.org/portal/js_pane/P-Help?URL=using.html#predict. Of course the input needs to have its stereochemistry defined by wedge bonds. Details on how to use the jars are found at http://sourceforge.net/p/nmrshiftdb2/wiki/PredictorJars/
(original date 2013-9-14) A short message only indirectly related to nmrshiftdb2 - I made the chemical database system behind nmrshiftdb2 a project in it's own, to be found on sourceforge.net. This can serve as a chemical database "cartridge", but works a bit differently from the common systems. It is possible to assign properties to bonds and atoms on the database level using referential integrities. If anybody is interested in using this, I would provide any assistence which might be necessary. The principles are tested, since used in nmrshiftdb2, but the build system of the new project might need some tuning.
(original date 2013-8-31) A did some brainstorming in nmrshiftdb2 software licencing (the licence for the data is a different thing). Until now, the software is licenced under an "artistic licence", which you can read here. This was chosen 11 years ago when the NMRShiftDB project started because it was considered a liberal licence. Whilst there have not been real problems so far, there are at least two potential pitfalls: a) This licence is not considered a free licence by the Free Software Foundation b) Integration of GPL Software is not possible.... read more
(original date 2013-2-13) Since last year, I did some work on removing duplicate structures from the database (the spectra were kept, of course, and all assigned to one structure). Just in case somebody wonders what happened so that duplicate structures occured, here is the main reason (the other reason are simple software bugs e. g. in SMILES generation or in the code doing the save to the database): nmrshiftdb2 requires information about which configurations around double bonds are actual E/Z-configurations as drawn and which are unspecified and just drawn randomly in one way (I still think that without this information, real decisions about structure identity cannot be made). In the past, this information was also asked in small rings where actual E/Z-configurations cannot occur. If in one case the bond was declared as "unspecified" and in the other as "as drawn", two entries with different IDs for the same structure were made. A typical example is molecule ID 2505 - this existed twice (the other ID was 20039308, but this has disappeared, really the only case where ID disappear) with the bond from 5 to 6 being considered in Z configuration in one of them, unspecified in the other. These have now been merged and 2505 has two spectra (both 13C here, but many of the merges lead to structures now having 13C and 1H spectra). 90 such merges happened.
(original date 2012-11-28) As announced earlier in the news, nmrshiftb2 is part of the Thomson Reuters Data Citation Indes (DCI). I did not yet have a chance to test it myself (no subscription at the local university), but I can show some screenshots showing the entry for nmrshiftdb2 (top) and two examples molecules with links to nmrshiftdb2. I think CDI is a good start to extend bibliographies beyond "old style" literature. Click on thumbnails below for example screenshots.... read more
(original date 2012-11-26) A new release was done over the weekend. The changelog gives details. I want to emphasize a bit the improved prediction. This is an extended HOSE code system, where chiral configurations are coded into the HOSE code. They are given relative to the chiral centre or the double bond in case of E/Z configurations, so it does not depend on 3d coordinates. I think this overcomes a major problem with schemes based on 3d coordinates, where it is difficult to establish the actual coordinates measured (of course it's easy to establish some coordinates). Examples of what the prediction can do are shown in the help. Of course quality depends on what is in the database and chiral centres must be specified by wedge bonds. So, what does this give altogether? Doing a 10 fold cross validation gives the following results for the old prediction: ... read more
(original date 2012-8-10) Over the past few days, I reworked the bibliographic data in nmrshiftdb2 for our new collaboration with Thomson-Reuters (see the nmrshiftdb2 news). Some of the older literature references were freetext fields and therefore not in a specific format. These have all been changed to the newer model where authors, title etc. are saved separately. Plus some obvious errors have been corrected. It is now possible to export in any desired format, e. g. bibtex. The file download at nmrshiftdb2 download page actually contains bibtex, so an automated processing is now possible. Of course this is not a guarantee that there are no typos etc., but it definitely is a step forward.
(original date 2012-6-28) Chemaxon recently announced an NMR predictor for their Marvin sketcher. This is based on nmrshiftdb2 data. I came across it at the really great User Group meeting in Budapest and I give my impressions here. Many thanks to Chemaxon for properly acknowledging nmrshiftdb2 and this criticism is meant as an encouragement, of course.
So Chemaxon, according to their presentation, took the nmrshiftdb2 data and calculated physicochemical and topological descriptors for the atoms. They use these to train multilinear least-squares regression (MLR) and support vector machine (SVM) models and do predictions based on these. So this is, for my understanding, an artificial intelligence approach.... read more
(original date 2012-6-28) This is used as a sort of blog by me (that is Stefan Kuhn). I will post occasional observations, remarks and thoughts about nmrshiftdb2 or its environment here. These should neither be news (which are on the nmrshiftdb home page) nor help topics (which are in the help pages) nor pure technical topics, which are in the wiki - really all the rest goes in here. I will see what this evolves into over time.... read more