Investigation of a charge-transfer substituent constant using computational chemistry and pattern recognition techniques

Abstract

Using the techniques of computational chemistry, a set of 58 parameters were calculated for 43 monosubstituted benzenes. Substituent constant values, κ, derived from the measurement of formation constants of charge-transfer complexes are available for 38 of these compounds. Relationships between κ and the calculated descriptors have been investigated using a variety of multivariate statistical techniques. Unsupervised analysis by two methods showed groupings of similar substituents and a tendency to order the compounds according to their κ values. It has been shown that κ can be predicted using these parameters but the different multivariate methods yielded different results. Partial least squares and principal components regression both tended to underpredict κ values whereas simple linear regression equations under- or over-predicted depending on the number of terms included. Examination of the descriptors involved in the correlations has shown that electronic effects are relatively unimportant in these complexes and that bulk and hydrophobicity parameters are most useful for the description of κ.