Optimising the EVA descriptor for prediction of biological activity

Abstract

EVA is a multivariate molecular descriptor for use in QSAR studies. It is constructed from vibrational eigenvalues derived from either a quantum theoretical or molecular mechanical treatment of molecular structure. This paper applies the method to biological-activity data using measures of the inotropic potential of a range of Calcium channel agonists. The performance of the descriptor, as both an explanatory and a predictive tool, is analysed in relation to the way in which it is constructed using a rigorous statistical treatment. Its capabilities are examined in relation to those of previously published methodology which used a composite descriptor. It is shown to have improved performance and several procedural advantages, such as ease of calculation and operation. It is a 3-D structural descriptor which does not require prior co-alignment of structures for a QSAR study.