Earlier this week we added a new capability to ChemSpider for our users. Using the web service provided via nmrdb.org we embedded the ability to predict an NMR spectrum from any record view on the ChemSpider website. The NMR prediction service is provided by Luc Patiny’s group out of Ecole Polytechnique Fédérale de Lausanne at the Institute of Chemical Sciences and Engineering. Their nmrdb.org webpage offers a series of services, not just NMR prediction and I offer the details below from their website.

NMR Predictor – This page allows to predict the spectrum from the chemical structure

NMR Assigner – Upload and assign NMR spectra on-line. The assignment of NMR spectra may be decomposed in 4 steps:

  1. identification of the signals
  2. integration and multiplicity determination
  3. assignment of each signal to the corresponding atom in the molecule
  4. exportation of the data for publication and/or for database storage

NMR Resurrector - A great amount of NMR information is currently available in the form of scientific publications. However, this information is not readily accessible in the format required for complex searches. The Resurrector enables the user to easily import these in-line spectral descriptions and creates an assigned visual representation that can be seamlessly integrated in the attribution process.

I am an NMR spectroscopist by training and have been involved with NMR either running NMR labs in academia, gov’t labs or Fortune 500 companies for almost a decade or involved with the development of commercial NMR software tools for prediction, processing and structure elucidation. Doing NMR prediction well is not easy. There are multiple approaches and many have been discussed previously on this blog so I won’t belabor that point. However, a set of online free utilities for prediction and assignment offers a new entry into the domain and the ease of integration allows anybody to connect up via their website in just a few minutes.

I haven’t had time to test the system rigorously on complex molecules but simple molecules look fine (based on a test set of about 5 molecules).

We have produced the integration in order to allow crowdsourced testing of the prediction algorithms. test it out. Provide the authors feedback as well as post your comments here. It’s easy to run…navigate to a record view of interest and look for the RED words “Predict NMR”. We will shortly provide you a way to predict the spectrum for any molecule via the ChemSpider structure input interface, it won’t have to be a part of our database.

Stumble it!

8 Responses to “NMR Prediction Now Available Via ChemSpider”

  1. Web as platform: Why I like web services #223 : business|bytes|genes|molecules says:

    [...] today I pointed to nmrdb.org on Friendfeed. A few minutes ago I saw a post by Antony Williams that uses the web services provided by nmrdb.org to provide NMR functionality on [...]

  2. David Bradley says:

    I wish stuff like this had been around when I was studying chemistry…would’ve made life so much easier ;-)


  3. hko says:

    H-NMR prediction of anisole.
    Seems to be a little bit odd to find five different
    shifts for the phenyl protons in anisole, if symmetry
    averaging is not taken into account.

  4. hko says:

    Prediction of 4-methoxybenzaldehyde.
    Sorry. Anisole was not a convincing example for missing
    symmetry averaging. See however predicted shifts for
    4-methoxybenzaldehyde. I can find four shifts at 7.957,
    7.966, 7.033, and 7.023 ppm for the phenyl protons.
    Is this due to missing symmetry averaging ? Calculated
    spectrum shows only a crude similarity compared to the
    real spectrum.

  5. hko says:

    H-Prediction of 1,4-dichlorobenzene.
    Should I forget my basic NMR-knowledge ???
    Calculating proton shifts of 1,4-dichlorbenzene,
    I found a multiplet pattern with EIGHT lines.
    That is more confusing than enlightening.

  6. hko says:

    Proton prediction of benzene.
    To the surprise of all ? 1H spectrum of
    benzene shows eighteen lines.

  7. Andrew Anderson says:

    Hi All,
    As someone who’s been using, then selling NMR prediction tools for quite a while now, I took great interest in the availability of online prediction through ChemSpider. As most of you know, there are quite a few commercially available prediction applications available on the market: ModGraph, Perch, ACD/Labs are commercial outfits that license NMR Prediction capabilities, to name just a few (full disclosure: I work for the latter :) ). So commercial implications of online prediction aside, I wanted to do a cursory comparison of results of a few simple structures to compare with when I’ve seen experimentally. I first put in para-Xylene, here’s the shifts that are returned:

    6.994 (1, 1H, ddd, J=7.936, J=1.630, J=1.198), 6.994 (2, 1H, ddd, J=7.936, J=1.630, J=1.198), 6.994 (4, 1H, ddd, J=7.936, J=1.630, J=1.198), 6.994 (5, 1H, ddd, J=7.936, J=1.630, J=1.198), 2.229 (7, 3H), 2.229 (8, 3H)

    So this doesn’t match what’s on the visual output (and I’m frankly not sure how to paste that into a blog comment)…perhaps that’s just a bug, or my misuse of the software.
    But, let me try to describe what I see as an output, from aromatic region upfield to 0ppm:
    1) A sort-of quartet ~7.015ppm
    2) Another complementary sort-of quartet at ~6.9985ppm
    3) A singlet at ~2.23ppm

    I looked up the reported chemical shift (notice I left it as singular, since I see these protons as degenerate, am I naïve?) for the aromatic protons. The reported shift (solvent dependent of course) ranges from 6.94ppm to 7.06ppm. A good friend of mine reminded me that you could see the long-range coupling between the methyl and aromatic protons. But what else?

    I suppose I want to make a point here:

    If the prediction can’t handle simple structures like para-Xylene, I’d shudder to think of its prediction performance on structures the discerning chemist would use to test its accuracy.

    So why the rant? There are three issues here:
    1) Sometimes I feel (and maybe I’m simply sensitive to it) that it’s assumed in some circles that commercial scientific software vendors are the “bad guys,” and that for whatever product there is on the market, there is a viable, free alternative. As I’m sure my counterparts at Modgraph and elsewhere would attest: we put a fair amount of money (time, resources, etc.) into our prediction applications. And although ACD/Labs and Modgraph have an ongoing debate as to which of us have a more accurate predictor, I’d say that these commercial results, while not free, provide reasonably accurate results. When and where we don’t provide accurate results, we strive to improve. Tony, you can attest to the hard work (10 years worth!) to get the ACD/Labs prediction applications to where they are today!
    2) I noticed an earlier comment to the effect of, “I wish this stuff would have been around when I was studying chemistry.” No slight to the commentor intended, but I’d personally find prediction results like these to be a detriment rather than a benefit.
    3) Tony, at what point will negative comments on the prediction capabilities you’ve exposed lead you to conclude it’s not ready for “prime time?” You and I both know very well that quality matters, and I believe these prediction capabilities do NOT meet your quality standards.

    I’m welcome to hearing feedback.

  8. Inchirieri Masini says:

    I used iwebtool for predictions.Works good

Leave a Reply