The information content of the eigenvalues from modified adjacency matrices: Large scale and small scale correlations

This analysis studies a type of chemical descriptor (BME) recently proposed by Burden [1], in which each molecule is identified by the eigenvalues of a matrix representing the hydrogen-suppressed connection table of the molec ule. Their ability to describe large noncongeneric databases and their suit ability for building QSAR models of chemical series were assessed. Our stra tegy has been twofold: a) we compared the information carried by the BME to that carried by several types of descriptors on a database of 112 nonconge neric chemicals representative of many chemical classes; b) we selected 15 QSAR models from the literature, relative to different properties/activitie s and solved through the use of different descriptors and approaches, and h ave made a comparison with models based on BME. Our analysis indicated that the information provided by many types of chemical descriptors (especially the topologically-based ones) are largely overlapping, and hence are quasi -equivalent for gross classifications (e.g. similarity searching, dissimila rity maximization). Regarding the analysis of individual chemical series, t he BME descriptors were adequate to build many different QSAR models, thus pointing to a remarkable versatility and interchangeability with other desc riptors (including complicated 3D-based descriptors like CoMFA). However, t his versatility is not absolute las shown by the BME failure to build 2 out 15 QSAR models). The implications of our findings for QSAR modelling are d iscussed.