NMRShiftDB at CHEMSPIDER

Since approximately 3 weeks (March 19^th, 2010 – at least at this time it came to my attention) CHEMSPIDER introduced a feature allowing visitors to inspect the C-NMR of their compounds. In cases where the identity is contained in the database, the experimental data from NMRShiftDB is displayed, in cases where the compound is not in the NMRShiftDB-database a spectrum prediction is launched and the predicted values can be highlighted by hovering with the mouse over the lines.

In principle, this is a nice feature – YES, really,

but why ^„in principle“ ?

The trick is simple – it's like in the supermarket. When you want to sell something, increase the price by let's say 20%, then you put a large poster showing the article and the message „Today only 40% off“ and within a short time this article is out of stock !

Its similar to using NMRShiftDB – the message here is 'Its free' ! Has anybody analyzed the quality of the data and the algorithms ? OK, you might remember the 'blog-war' I have initiated about 2 years ago, when I analyzed the quality of the underlying data and stated that the quality was poor. I agree, there was a lot of effort to curate these data and the data quality has become much better (BTW: Also with the help of ACD & CSEARCH; I have done many predictions/checks for the main-contributor 'hko', because we have a good personal relationship and I highly appreciate his feedback with respect to CSEARCH)

Now let's go into a more detailed analysis

of the algorithms in NMRShiftDB:

Lets take the compound:

http://www.chemspider.com/Chemical-Structure.63280.html

1-Bromo-3-methyl-but-2-ene: you get - as is to be expected – two signals for the methylgroups, the cis-Me is at 17.6 ppm, the trans-Me at 25.9 ppm.

Now lets change the Br against a PH2-group – the data can be found in J.Org.Chem.,63,59(1998) for this compound. It is also in Chemspider:

http://www.chemspider.com/Chemical-Structure.9019638.html

Now a C-NMR spectrum must be predicted, because this compound is not in the database of NMRShiftDB.

What happens ?

We end up with ONE, SINGLE signal at 20.3ppm with an intensity of 2 carbons for both methyls. You can substitute the 'PH2' by any other substituent, as long as a prediction is necessary you end with ONE single signal for the cis- and trans-methyl group. In 1978 Wolfgang Bremser at BASF could already separate the 2 distinct ranges in the case of cis/trans-isomerism.

Conclusion: 20.3 ppm seems to be the mean value of all cis- and trans-methyl groups in the NMRShiftDB-database, but I want a value for the cis-Me and the trans-Me. It's like putting your left foot into boiling water, your right foot into an ice-bath – on average you feel well ! Mean values are meaningless !

Therefore my question:

Why do Christoph Steinbeck and Stefan Kuhn bother the chemical community with technology from more than 30 years ago !

(YES – It's free, I forgot !)

Another interesting example:

Take the structures 6413 / (3b,5a-isomer); 59456 (3a,5a-isomer); 191826(3b,3b-isomer) and get the C-NMR data, focus only on C3, the most downfield one – its always at 72.5ppm totally independant of the configuration at C-3 and totally independent of the cis/trans ring-junction between A&B-ring in a steroid.

Thats great, C-NMR is not a good method to

differentiate between diastereomers !

( I have learned that from NMRShiftDB – it must be correct ! )

I repeat my question:

Why do Christoph Steinbeck and Stefan Kuhn bother the chemical community with technology from more than 30 years ago !

(YES – It's free, I forgot!)

I have written a paper with the title „Do high-quality ¹³C-NMR spectral data really come from journals with high Impact Factors“ [W. Robien, TrAC, 28, 914 (2009)] – maybe we will see a significant change in the quality of published NMR-data when a lot of people rely for their assignments and structure elucidations on predictions done by NMRShiftDB.

Good night, Community !

Examples:

Geraniol:

Example for Me-cis/Me-trans: Geraniol is contained in both datasets used, CSEARCH can make use of the stereochemical information. When artificially downgraded to the functionality of NMRShiftDB ignoring stereochemical information, then large deviations at BOTH methylgroups appear. Click here

Cholestanol:

Cholestanol: 4 Stereoisomers. NMRShiftDB produces always 71.3ppm for C3 independant of configuration at C3 and C5.

Isomer #1

Isomer #2

Isomer #3

Isomer #4

Now you might argue, that in the second and third line the „NMRShiftDB“-prediction is by 0.2ppm or 0.9ppm respectively, better. You are absolutely right, a defect watch gives you twice a day the correct time – the problem is, you don't know when !

Alkyl-chains:

n-Alkyl-chains are known to give the following signals: There is a quartett around 14ppm, and 3 triplets around 22ppm, 29.5 ppm and 31ppm

Let's take hexylbenzene as an example:

You find 20.3 ppm and 5 x 29.6 ppm for the n-hexyl-chain.

We compare here the CNMR-data of

n-decane (C₁₀H₂₂) and n-nonadecane (C₁₉H₄₀):

For n-decane 5 signals are shown at 14, 23, 32 and around 30ppm (2 signals) as to expected – this compound is contained in the database of NMRShiftDB, therefore this display is based on recalling the shiftvalues from the internal database – it has nothing to do with SPECTRUM PREDICTION !

Now let's do the same for n-nonadecane:

In this case we get a „spectrum-prediction“ containing 2 signals, one around 20ppm corresponding to 2 carbons and another signal at 30ppm corresponding to 17 carbons.

I think these examples are self-explanatory with respect to the quality of a C13-spectrum prediction based on NMRShiftDB.

If anybody wants to start another 'blog-war' – please keep in mind:

I have done a bulk-order of kevlar-vests two years ago

A lot of NMR-textbooks have to be rewritten to take into account, that stereochemistry has no influence on carbon chemical shift values.

A lot of NMR-textbooks have to be rewritten to take into account,

that every n-alkylchain has a quartet at 20ppm and

all other CH₂-groups are around 30ppm.

In order to get a state-of-the-art spectrum prediction

use the interface at

http://nmrpredict.orc.univie.ac.at/c13robot/robot.php

or

http://synthon.pch.univie.ac.at/csearchlite/robot.php

(3 predictions per day are free of charge – I reserve the right to remove this tool without announcement whenever it seems to be necessary)

Page written by: Wolfgang.Robien(at)Univie.ac.at

Page online since: April 8^th, 2010