A combined semiempirical MO neural net technique for estimating C-13 chemical shifts

A back-propagation artificial neural net has been trained to estimate C-13 chemical shifts from the results of AM1 and PM3 semiempirical MO calculations. The input descriptors include the atom-centered monopole, dipole and quadrupole moments derived from the natural atomic orbital/point charge (NAO/PC) model, the four highest bond orders to the carbon atom being considered and the elements to which these bonds are made. The resulting net estimates the chemical shifts of a test set of 156 chemical shifts with a standard deviation of less than 7 ppm from the experimental values for AMI and slightly more for PM3.