Home   |   Products   |   Consulting Services   |   Contact us    

 

NMRPredict:

  Modgraph Home
  NMRPredict Overview
  Carbon 13 NMR Prediction
  Proton NMR Prediction
  NMRPredict version 3
  NMRPredict version 2
  Try NMRPredict
  NMRPredict and Mestrelab Research
  NMRBenefit
  NMRPredict Help Files
  Pricing

 

 

 

NMRPredict C13 Prediction Accuracy

In May 2007 ACDLabs made a comparison of their ACD/CNMR Predictor with NMRPredict using the 214,136 assigned carbon chemical shifts in NMRShiftDB. This was published both on their website and in a printed document. They claimed that “ACD/CNMR Predictor vastly outperforms the prediction accuracy of the CSEARCH (Modgraph) algorithm.  The average deviation in ACD/CNMR predictor was 1.59 ppm compared to an average deviation of 2.2 ppm in CSEARCH (NMRPredict)”.

Their claim was based on the fact that Dr Robien recently posted a page on his website where he looked at the data quality of the free C13 NMR spectra available in NMRShiftDB.  In his article he mentioned an average deviation of 2.22ppm for his program when used with the NMRShiftDB database.

What ACD failed to notice was that Dr Robien was never intending to prove how accurate his prediction algorithms were in his original article. The intention of his article, which was stated very clearly, was to demonstrate how many errors could easily be found in an un-verified database such as NMRShiftDB. A detailed description of Robien's parameter selection can be found here.

For his article Dr Robien used a Neural Network developed in 1996 based on only 60,000 records. NMRPredict uses both a HOSE code database of 345,000 C13 records and a Neural Network at the same time. Furthermore it includes a "Decision-Engine" which selects for a particular carbon atom, which value should be taken – either the HOSE-code prediction or the prediction coming from the Neural Network. This selection process - leading to a "BEST-VALUE" - is based on many statistical parameters coming from both prediction technologies and some rule-based criteria, with the effect of dramatically improving the final result.

We re-ran the predictions of the NMRShiftDB data using NMRPredict and its 'BEST' technology. We received an overall average deviation of 1.40 ppm. This result outperformed the results of 1.59 ppm published using the ACD/CNMR Predictor software version 10.05. A detailed report about how we ran our evaluation is also available.

Of course the only way to really judge the accuracy of a prediction program is to test it with your own molecules. We encourage you to try NMRPredict yourself. Go to the Mestrelab Research website and either download Mnova with the included NMRPredict Desktop plug-in or download the NMRPredict client and try your predictions on-line.

Alternatively if you would like a 45 day, free-of charge, in-house trial of NMRPredict please contact Jeff Seymour at jeff@modgraph.co.uk. for more details.