GRid-INdependent Descriptors (GRIND): A Novel Class of Alignment-Independent Three-Dimensional Molecular Descriptors

Top Cited Papers
Publisher Website
Google Scholar
Abstract
Traditional methods for performing 3D-QSAR rely upon an alignment step that is often time-consuming and can introduce user bias, the resultant model being dependent upon and sensitive to the alignment used. There are several methods which overcome this problem, but in general the necessary transformations prevent a simple interpretation of the resultant models in the original descriptor space (i.e. 3D molecular coordinates). Here we present a novel class of molecular descriptors which we have termed GRid-INdependent Descriptors (GRIND). They are derived in such a way as to be highly relevant for describing biological properties of compounds while being alignment-independent, chemically interpretable, and easy to compute. GRIND are obtained starting from a set of molecular interaction fields, computed by the program GRID or by other programs. The procedure for computing the descriptors involves a first step, in which the fields are simplified, and a second step, in which the results are encoded into alignment-independent variables using a particular type of autocorrelation transform. The molecular descriptors so obtained can be used to obtain graphical diagrams called “correlograms” and can be used in different chemometric analyses, such as principal component analysis or partial least-squares. An important feature of GRIND is that, with the use of appropriate software, the original descriptors (molecular interaction fields) can be regenerated from the autocorrelation transform and, thus, the results of the analysis represented graphically, together with the original molecular structures, in 3D plots. In this respect, the article introduces the program ALMOND, a software package developed in our group for the computation, analysis, and interpretation of GRIND. The use of the methodology is illustrated using some examples from the field of 3D-QSAR. Highly predictive and interpretable models are obtained showing the promising potential of the novel descriptors in drug design.

This publication has 9 references indexed in Scilit: