Putting ACD/NMR Predictors
to the Test

During the history of the development of the ACD/Labs NMR prediction tools we have compiled and checked data from thousands of literature articles to build databases of over 165,000 assigned structures for ¹H and ¹³C, 8000 for ¹⁵N, 13,800 for ¹⁹F, and 22,600 for ³¹P. With these assignments and structures as a basis and using our correlation algorithms developed over the past ten years, it should be obvious that our fragment-based predictions will offer superior performance to any rules-based systems. Fragment-based prediction offers the opportunity to cover wide ranges of structural diversity not available to rules-based systems. It is this type of performance that has enabled Advanced Chemistry Development to become the industry leader in NMR prediction today. Rigorous testing of our capabilities relative to other software packages has resulted in our software becoming the package of choice on a worldwide level for companies such as Pfizer, Astra Zeneca, GlaxoSmithKline, 3M, Eastman Kodak Co., Millennium Pharmaceuticals and many others.

There have been previous reports of our capabilities relative to other software packages performing NMR prediction today. Recently, we have reported an unprecedented study to compare the prediction accuracy of ACD/HNMR and CNMR (version 8.0, from December 2, 2003) with CambridgeSoft ChemNMR (version 8.0, from April 23, 2003). A dataset was created consisting of more than 120,000 ¹H and ¹³C NMR chemical shifts extracted from recently published structures and chemical shift assignments. Each program was used to predict the chemical shifts for each of the structures in the dataset and then compared these values to the experimental values extracted from the literature.

We report here our results with version 8.0 of our software as an indication of present capability. During this comparison it is notable that ACD/NMR Predictors offer many capabilities that are unique to our software. These include:

Numerical value for the prediction error estimate (confidence limit)

• The ability to predict spectra for different resonant frequencies
• Solvent specific prediction
• A database of chemical shifts
• The ability to train the predictions with new experimental data
• Prediction of coupling constants
• The ability to predict 3J H-H coupling constants using 3D optimized structures and the Karplus equation
• Visual identification of exchangeable protons on the display
• An interactive display of predicted spectra
• Actively linked objects in the display (Spectrum, table of shifts and the table of coupling constants)

Since, as chemists, we are accustomed to looking directly at chemical structures with real experimental spectra (with multiplets, real line widths etc.) rather than stick diagrams, we believe the interface components for spectral display and manipulation are crucial in any spectral prediction tool. For proton this includes the ability to display integrals, interaction between coupling groups and complex multiplets associated with a single or multiple protons. For carbon the ability to display J-modulated spectra and perform DEPT type spectral editing is of value. All of these are possible in the ACD/NMR spectral prediction interface. An ideal way to see this strength is to view the demo movies but an example display screen is shown below.

Even though interface strength is certainly very powerful and important, the ability to perform good predictions is of course the first priority. Below we show comparisons of Cambridgesoft/ChemNMR to our Version 8.0 releases. In the present study, the error of prediction was calculated by the following formula for each program as shown in the table below.

Standard Error (ppm) =

There σ is Standard Deviation of the Predicted from Experimental shifts and n is the number of shifts in the dataset.

¹H NMR Prediction Comparison

Table 1: Statistics comparing the accuracy of ACD/HNMR predictions to CambridgeSoft ChemNMR.

The table above illustrates that ACD/HNMR has an error of about half that of ChemNMR using the same dataset. In addition, over 62% of our predictions are within 0.1 ppm of the experimental chemical shift as opposed to only 27% in ChemNMR (Figure 1).

Figure 1: ¹H NMR prediction comparison based on the difference between the predicted and experimental NMR shift. Note that 63% of the shifts predicted in ACD/HNMR were within 0.1 ppm.

¹³C NMR Prediction Comparison

Table 2: Statisics comparing the accuracy of ACD/CNMR predictions to Cambridge Soft

Once again the table illustrates the remarkable accuracy advantage of ACD/CNMR as its error is less than half that of ChemNMR. Fully 64% of the predictions from CNMR are within 1 ppm of the experimental shift as opposed to only 32% in ChemNMR (Figure 2).

Figure 2: ¹³C NMR prediction comparison based on the difference between the predicted and experimental NMR shift. Note that about 64% of the shifts predicted in ACD/CNMR were within 1 ppm.

Please request the complete comparison document from your account manager or email sales@acdlabs.com.