Statistical Approaches to Estimating the Relative Contribution of Intermolecular Interactions in Aliphatic Alcohols: Application to QSPR/QSAR Modeling of Their Boiling Points

Abstract

A three-layer feed forward neural network trained with a Levenberg–Marquardt batch error back propagation algorithm has been used to model the strong relationships between the boiling point of aliphatic alcohols and intermolecular forces consisting both in Van Der Waals forces and polar interactions, respectively. For that purpose, we use the multifunctional autocorrelation method to provide an appropriate topological description. Two types of descriptors are generated: the first is commonly used in QSARs and QSPRs modelling, it gives a general description of the whole of the molecule; the second is attributed to the local description of the group hydroxyl. In this we have turned our interests to the explanatory capacities of our methodology to explore the relative contribution and/or the contribution profile of the input factors compared to the size of the molecule. The initial data set is divided into different subsets in increasing order of values of boiling point. Then, we explore the good descriptive ability of the molecular descriptors calculated solely from the modified autocorrelation method to carry out a variable analysis and give information about features of the compounds responsible for their boiling points. This is made possible by comparing the regression coefficients for the established linear model, and by using the Garson weight portioning method for the ANN analysis.