Abstract

Three-way analyses of quantum topological molecular similarity descriptors were used for quantitative structure property relationship modeling of the acidity constant of some phenol derivatives. A three-way data was built for different molecules by constructing a data matrix for each molecule. The matrix was produced by considering different bonds in each molecule and different descriptors in each bond. The three-way models parallel factor analysis and N-way partial least squares, and two way models including partial least squares were used for modeling structure–acidity relationships. Comparison of the results showed that the three way arrays produced more predictive models with lower over-fitting. The bilinear partial least square model resulted in a biased estimation of acidity constant of prediction set with average relative error of prediction of 1.87%, whereas that obtained by parallel factor analysis and N-way partial least squares was near to zero (i.e. )0.41 and )0.33, respectively). Additionally, the three-way methods allowed investigating the significant impact of different bonds and different descriptors using leverages of the parallel factor analysis loadings.