NMRPredict Help

		Help overview
		Running a Prediction
		C13 NMR Prediction in Detail
		Proton NMR Prediction in Detail

How do we calculate the "Best" proton prediction value?

The "Best" proton predicted values are calculated by using statistical algorithms developed by Professor Robien. In 2007 we were given access to 90,000 assigned proton spectra containing over 1,000,000 shift values. We predicted every one of these atoms and compared them with the native values predicted by the conformer and increment approaches. What we discovered was that there was no simple rule which we could apply to say that one method should be used rather than the other for a specific chemical class.

We decided to break the 1,000,000 shift values down into 6,500 functional groups and develop a routine for each group which said what "correction factor" should be applied to the two native values from Ernö Pretsch's increment approach and Ray Abraham's conformer method. We then apply a weighted average to the two methods. The correction therefore does the following: Add or subtract x to the native values from the two methods and then take a weighted average of the two methods.

If in the "From" column you see "f(C,I)" it means that the functional group around the atom has been seen in the 6,500 training set and a correction factor has been applied.

What happens if we cannot apply a correction factor?

Of course 6,500 functional groups do not cover all of chemistry and sometimes we cannot apply a correction factor. In these cases the "Best" value will simply be the native value from the increment method. This value was chosen rather than the conformer value because statistically it was slightly more accurate. In the "From" column you will see "=I". However if the increment method reported an error in the "Info" column (which means that the native increment value may be inaccurate) we take the native conformer value and "=C" will be seen in the "From" column.

If we have been unable to apply a correction factor the values will be shown in red in the Overview tab. It is recommended that users assign values to their own databases where we cannot apply a correction.

In the example of strychnine below it can be seen that atoms 9 to 19 can have a correction factor applied. All other atoms have not been seen by the 6,500 functional group training set, no correction can be applied and all we can do is provide the native value from the increment method.