On Characterization of Heteroatoms

In summary, one may say that in regression analyses heteroatoms need to be differentiated from carbon atoms as has been shown by Kier and Hall 40 years ago, but instead of prescribing a single set of modified indices for any individual heteroatoms, one should consider two or more such sets of descriptors. Two heteroatom descriptors, when combined, will cover a wider structure space territory than a single index having preselected weight to represent heteroatoms covers even if helped by the plain connectivity index x- That indeed this happened we have seen from a few regressions reported in the book by Kier and Hall. 

Very recent and quite independent support for this comes from a study of Pogliani and Julian-Ortiz on QSPR with descriptors based on averages of vertex invariants [107]. They introduced a new type of indices, the mean molecular connectivity indices, based on a number of different concepts of mean. They have been interested in comparative study of multiple regression analysis (MRA) and artificial neural network (AAN) approaches. For illustration, they examined a selection of physicochemical properties of organic solvents, using the mean molecular connectivity 

---