Toward More Reliable ¹³C and ¹H Chemical Shift Prediction:  A Systematic Comparison of Neural-Network and Least-Squares Regression Based Approaches

Abstract

The efficacy of neural network (NN) and partial least-squares (PLS) methods is compared for the prediction of NMR chemical shifts for both 1H and 13C nuclei using very large databases containing millions of chemical shifts. The chemical structure description scheme used in this work is based on individual atoms rather than functional groups. The performances of each of the methods were optimized in a systematic manner described in this work. Both of the methods, least-squares and neural network analyses, produce results of a very similar quality, but the least-squares algorithm is approximately 2--3 times faster.

A systematic approach for the generation and verification of structural hypotheses

“…For each atom within the candidate structure the related structures used for the prediction can be shown with their assigned chemical shifts and this allows the user to understand the origin of the predicted chemical shifts. All three methods can be used for 1 H, 13 C, 15 N, 19 F and 31 P NMR chemical shift prediction and all of them are implemented within the Structure Elucidator software program.…”

“…For the purpose of structure determination and confirmation, 1D NMR spectra ( 1 H, 13 C, 15 N and 19 F) in combination with 1 H-13 C gs-HMBC and 1 H-15 N gs-HMBC data were used. The authors [30] derived the two pairs of alternative isomers shown above and the problem was reduced to selection of the correct structure within each of the isomer pairs.…”

The application of empirical methods of 13C NMR chemical shift prediction as a filter for determining possible relative stereochemistry

“…The fastest NMR spectra calculations are provided using an incremental approach and offer a computational speed of 6000-10 000 chemical shifts per second on a normal desktop computer (ca 2007) and provides an average chemical shift deviation for carbon NMR of 1.8 ppm. [1,2] Spectral prediction utilizing artificial neural networks provides similar speed and accuracy performance. [1,2] The third most popular empirical method is slower and is based on the application of a database containing the reference structures with assigned 13 C or 1 H chemical shifts.…”

“…[1,2] Spectral prediction utilizing artificial neural networks provides similar speed and accuracy performance. [1,2] The third most popular empirical method is slower and is based on the application of a database containing the reference structures with assigned 13 C or 1 H chemical shifts. The target and reference structures are described by means of HOSE codes [3] and this allows prediction of the chemical shift of an atom from the target structure using the chemical shifts of the reference structures as the basis.…”

“…The incremental and neural nets based algorithms of chemical shift prediction also use the stereochemistry information related to the atoms included into 3-6-membered cycles. [2] It was interesting to know whether this information can be useful for stereochemistry determination.…”