Local Descriptors for Nuclear Magnetic Resonance Spectroscopy

2 Local Descriptors for Nuclear Magnetic Resonance Spectroscopy 

The prediction of chemical shifts in H-NMR spectroscopy is usually more problematic than in C-NMR. Experimental conditions can have an influence on the chemical shifts in H-NMR spectroscopy and structural effects are difficult to estimate. In particular, stereochemistry and 3D effects have been addressed in the context of empirical H-NMR chemical shift prediction only in a few specific situations [81,82]. Most of the available databases lack stereochemical labeling, assignments for diastereo-topic protons, and suitable representations for the 3D environment of hydrogen nuclei [83]. This is the point where local RDF descriptors seemed to be a promising tool. 

The approach presented here uses a combination of physicochemical, topological, and geometric information [84]. The geometric information is based on local proton RDF descriptors to characterize the chemical environment of the proton. CPG neural networks established the relationship between protons and their H-NMR chemical shifts. Four different types of protons were treated separately regarding their chemical environment protons belonging to aromatic systems, nonaromatic 7t-systems, rigid aliphatic substructures, and nonrigid aliphatic substructures. Each proton was represented by a fixed number of descriptors. 

The mathematical flexibility of RDF descriptors and its use for specific representations has been mentioned before. The descriptors used in this approach have been 

Geometric descriptors were based on local RDF descriptors (see Equation 5.20) for the proton j 

---