Home   |   Products   |   Consulting Services   |   Contact us    



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




NMRPredict versus ACD CNMR/Predictor

ACD published on its website a paper named "NMR Prediction Accuracy Validation" , where they claimed to compare their CNMR/Predictor with our NMRPredict software. This "comparison" was based on a webpage published by Wolfgang Robien , where he obtained a deviation of 2.22/2.19 ppm using the NMRShiftDB-dataset . Wolfgang Robien clearly stated on his webpage, that this result was obtained using his CSEARCH program. This can be easily seen from the embedded screenshots.

NMRPredict uses within its prediction engines - based on HOSE-code and Neural Network technology - the same algorithms which are in use within CSEARCH. However, thanks to the collaboration between Wolfgang Robien and our development team, a few dramatic improvements have been implemented into NMRPredict.

NMRPredict and CSEARCH:

  • As Wolfgang Robien stated clearly on his webpage, CSEARCH is his environment for developing further pieces of software and his environment for generating high-quality databases which have gone through the most thorough data evaluation algorithms available
  • NMRPredict uses the same prediction engines as used in CSEARCH
  • NMRPredict has an engine built-in which automatically decides, from the query structure, if the calculations should be performed utilizing the stereochemical information
  • NMRPredict performs, in a different way from CSEARCH, solvent-specific predictions which leads to improved results
  • Both above mentioned features automatically optimize a few switches, which are passed to the prediction engines. No changes to the core-algorithms are made
  • When both predictions have finished, NMRPredict automatically decides which of the HOSE-code or the Neural Network based values should be regarded as the 'BEST-value'. This decision engine is based on many statistical parameters coming back from the prediction engines and some basic rule-driven setups
  • Now the results:

    Reference Data Set used: NMRShiftDB
    Technology used: NMRPredict BEST-technology
    NMRPredict databases used: ABCDGH - corresponding to 345,308 entries
    Stereochemistry settings: Auto-stereo recognition
    Solvent settings: All solvents used

    1.40 ppm overall average deviation

    ACD CNMR Predictor 1.59 ppm overall average deviation

    For this evaluation the combined databases from CSEARCH and SPECINFO holding a total of 345,308 reference spectra were used. Based on this higher number of reference spectra a somewhat higher structural overlap between our databases and the NMRShiftDB-testdata has been detected. In order to compensate for this, we have recalculated our overall average deviation of 1.40 ppm using the lower structural overlap as detected by ACD. The value of 1.40 ppm corresponds to 92,927 known carbon environments and 121,209 unknown carbon environments - without this compensation our overall average deviation would be slightly better, but a comparison with ACD's results would be impossible.