Carbon-13 nuclear magnetic resonance spectrum simulation

The 13 C NMR chemical shift of sp 3 carbon atoms situated in the R position relative to the double bond in acyclic alkenes was estimated with multilayer feedforward artificial neural networks (ANNs) and multilinear regression (MLR), using as structural descriptors a topo-stereochemical code which characterizes the environment of the resonating carbon atom. The predictive ability of the two models was tested by the leave-20%-out cross-validation method. The neural model provides better results than the MLR model both in calibration and in cross-validation, demonstrating that there exists a nonlinear relationship between the structural descriptors and the investigated 13 C NMR chemical shift and that the neural model is capable to capture such a relationship in a simple and effective way. A comparison between a general model for the estimation of the 13 C NMR chemical shift and the ANN model indicates that general models are outperformed by more specific models, and in order to improve the predictions a possible way is to develop environment-specific models. The approach proposed in this paper can be used in automated spectra interpretation or computer-assisted structure elucidation to constrain the number of possible candidates generated from the experimental spectra.

Section: Introductionmentioning

confidence: 99%

¹³C NMR Chemical Shift Prediction of the sp³ Carbon Atoms in the α Position Relative to the Double Bond in Acyclic Alkenes

Ivanciuc

Rabine

Cabrol‐Bass

et al. 1997

“…Two practical approaches for estimating the chemical shift of a carbon atom are direct database retrieval methods [1][2][3][4][5][6] and empirical modeling techniques. [7][8][9][10][11][12] The direct retrieval approach makes use of a database of chemical structures and their corresponding experimentally measured and assigned 13 C NMR spectra. Every carbon atom in each structure is represented algorithmically in some manner that relates its chemical environment to its chemical shift.…”

Section: Introductionmentioning

confidence: 99%

Automated Spectrum Simulation Methods for Carbon-13 Nuclear Magnetic Resonance Spectroscopy Based on Database Retrieval and Model-Building Strategies

Schweitzer

Small

1997

An automated method for the prediction of carbon-13 nuclear magnetic resonance ( 13 C NMR) chemical shifts which involves the combined use of database retrieval and multivariate calibration methods is described. A commercial database of chemical structures and experimentally measured and assigned 13 C NMR spectra is used to implement this methodology. The chemical environment of each of the carbons in the database is encoded as a vector, and Euclidean distance comparisons are made between these vectors and the similarly encoded environment of a target carbon whose chemical shift is to be predicted. This procedure yields a calibration set of carbon atoms with chemical environments similar to that of the target carbon. If the environment of the closest match is very similar to that of the target carbon, the predicted chemical shift is assigned as the chemical shift of that closest match (i.e., a direct chemical shift retrieval is performed). Otherwise, a chemical shift model based on spectra-structure relationships is built using the selected calibration set of carbons. The predicted chemical shift is then calculated with the computed model. The independent variables in the model are derived from numerical structural descriptors which describe some topological, geometrical, or electronic aspect of the chemical environment of the carbon atoms in the calibration set. Multiple linear regression, partial least-squares regression, and principal component regression are compared in terms of their effectiveness for use in building the chemical shift models. A series of experiments is performed to test the accuracy of direct chemical shift retrieval versus model building and to determine the optimal settings of several parameters that affect the model-building step. Based on a test set of 38 504 carbons, an overall mean deviation of 1.85 ppm between predicted and actual chemical shifts is achieved by use of direct retrieval alone, while the corresponding mean deviation based on a combination of direct retrieval and model building is 1.69 ppm.

“…Two practical methods for chemical shift prediction are direct database retrieval methods [1][2][3][4][5][6] and empirical modeling methods. [7][8][9][10][11] The direct database retrieval method is dependent on the availability of a large database of spectra and structures in which a structural representation of each carbon atom in the database is stored along with its experimentally measured chemical shift. The structural representation requires an algorithm in which the chemical environment of the carbon is encoded in a manner that correlates with its chemical shift.…”

Section: Introductionmentioning

confidence: 99%

Enhanced Structural Encoding Algorithm for Database Retrievals of Carbon-13 Nuclear Magnetic Resonance Chemical Shifts

and

Small

1996

An enhanced version of an algorithm is discussed which encodes a description of the chemical environment of carbon atoms in a manner that correlates to carbon-13 nuclear magnetic resonance (13C NMR) chemical shifts. The encoding algorithm uses a vector-based approach in which the first dimension of the vector represents the chemical shift of the carbon atom, the second dimension represents the collective influence of atoms one bond away from the carbon on its chemical shift, and each successive dimension represents the influence of the atoms one bond further away. This encoding algorithm is a key component of a 13C NMR spectrum simulation procedure in which each of the carbons in a large database of known structures and spectra is represented as a vector. Database search methods based on vector comparisons are used to find the closest matching chemical environments and associated chemical shifts for each of the carbons in a structure input by a user. Enhancements to the original algorithm include an expansion of the number of atom classes treated, the addition of a scheme to treat aromatic systems as a special case, and the use of an expanded vector format to regain some of the information lost by collapsing the molecular structure to a vector representation. To test this algorithm, a database of structures and spectra is split into training and test sets consisting of 16 959 and 4240 structures, respectively. Experiments performed to optimize several parameters associated with the encoding algorithm are followed by comparing the retrieved (i.e., predicted) and actual chemical shifts for the structures in the test set. For the optimal parameter settings found, the median of the mean absolute deviations in chemical shifts for the structures in the test set was 1.30 ppm and was obtained with an expanded vector representation based on 15 dimensions.