Home   |   Products   |   Consulting Services   |   Contact us    



  Chemical Registration
  Reaction-based Libraries
  Chemical Supplier Catalogue
  Database Integration
  Structure Drawing Control
  Imaging Suite
  Barcode Solutions




Press Release

January 2008

MODGRAPH are delighted to announce a hugely significant improvement in proton prediction. Recent developments have resulted in an improvement of 40% in the proton prediction accuracy in Modgraph's NMRPredict program!

Modgraph and their partners in proton NMR prediction, Professors Ernö Pretsch and Ray Abraham, have always hoped and strived for proton prediction accuracy of between 0.1 and 0.2 ppm average deviation.

Over the past few years many customers have tested NMRPredict and compared it to other proton prediction programs on the market. Their findings were always very similar - they reported that all programs gave an average deviation of 0.3 ppm at best. An improvement was clearly required.

Thanks to Modgraph's relationship with Wiley, Modgraph have recently gained access to 90,000 assigned proton spectra, which contained over 1.1 million shift values of protons attached to carbon. In August Modgraph predicted all 90,000 structures and evaluated the results. At that time Professor Pretsch's program gave an average deviation of 0.30 ppm and Professor Abraham's program gave an average deviation of 0.28 ppm. Most significantly, the "Best" prediction, if we were able to choose the best prediction on an atom-by atom basis, was 0.15 ppm! This was a hugely significant finding. It proved that the two prediction methods were very complimentary. The question then was could we develop a method to automatically select the "Best" value from the two predictions?

Wolfgang Robien has been working on statistical methods for many years which we believed could be used to make this "Best" selection. In October he delivered a program to us which gave us an automatic "Best" selection of 0.22 ppm on the 90,000 structures. Later that month Professor Pretsch gave us a new version of his increment prediction program which now makes use of 3D structures for the first time. We found that this alone brought his prediction accuracy down to 0.21 ppm, the "Best Possible" prediction down to 0.14 ppm and our "Best" prediction down to 0.18 ppm.

A summary is:

  • 0.30 ppm - Professor Pretsch's original prediction alone
  • 0.28 ppm - Professor Abraham's prediction alone
  • 0.21 ppm - Professor Pretsch's prediction alone once he took 3D molecules into account
  • 0.14 ppm - The "Best Possible" prediction if we could always select between the two methods
  • 0.18 ppm - Our current "Best" value

Our work has not finished. We are already in the process of re-running the 90,000 molecules with different settings to see whether we can further improve the prediction accuracy. We are confident that further improvements will come soon.

Please note that the 0.18 ppm accuracy we quote here RELATES TO THE PREDICTION OF THE 90,000 WILEY STRUCTURES. We do not claim that NMRPredict will always be able to reach a prediction accuracy of 0.18 ppm average deviation on your data. However, we do believe that the 1.1 million shift value were diverse enough, and the methods we are using to select the "Best" value are flexible enough, that most customers should see a similar improvement in prediction accuracy in NMRPredict and we are confident that the proton NMR predictions in the program will be significantly more accurate than those found in any other program.

The only way to test this claim is for you to test NMRPredict yourself. We suggest that this is best tested by downloading the new free 45 day trial of Mestrelab Research's Mnova including NMRPredict Desktop.