Quantitative molecular similarity methods in the property/toxicity estimation of chemicals: A comparison of arbitrary versus tailored similarity spaces

Quantitative molecular similarity analysis (QMSA) methods based on structure spaces using a variety of calculated molecular descriptors and experimental properties have been used for selection of analogs and estimation of a variety of properties pertaining to chemistry, drug discovery, and environmental toxicology. The set of descriptors, orthogonalized or not, have been selected either arbitrarily or based on some intuitive notion of their association to the property of interest. In this study we have developed a method of 'tailoring' the QMSA method to a particular property of interest through the use of indices which are strongly correlated with the property instead of using an arbitrary set of parameters. Comparative studies of the tailored vis-à-vis two non-tailored (arbitrary) similarity spaces based on principal components derived from topological indices show that the tailored method outperforms the two non-tailored techniques in the QMSA-based estimation of hydrophobicity (log P, octanol/water) for a set of 213 compounds and Ames' mutagenicity of a group of 95 aromatic and heteroaromatic amines.