When QSAR models are fitted, it is important to validate any fitted model-to check that it is plausible that its predictions will carry over to fresh data not used in the model fitting exercise. There are two standard ways of doing this-using a separate hold-out test sample and the computationally much more burdensome leave-one-out cross-validation in which the entire pool of available compounds is used both to fit the model and to assess its validity. We show by theoretical argument and empiric study of a large QSAR data set that when the available sample size is small-in the dozens or scores rather than the hundreds, holding a portion of it back for testing is wasteful, and that it is much better to use cross-validation, but ensure that this is done properly.

For correlating selected properties of amino acids we use the variable connectivity index 1χf which is obtained by generalizing the connectivity index 1χ by introducing variable weights. The generalized connectivity index assigns to each atom a parameter yet to be determined during the regression analysis so that the standard error of estimation becomes as small as possible. Using the weights −0.65, 5.25, 0.20, and 0.50 for carbon, oxygen, nitrogen, and sulfur, respectively, we obtained a multivariate regression with the regression coefficient 0.9865, standard error of 4.73, and Fisher ratio F=506 for partial molar volumes of amino acids. © 2000 John Wiley & Sons, Inc. Int J Quant Chem 80: 1199–1209, 2000

Due to the lack of experimental data, there has been increasing use of theoretical structural descriptors in the hazard assessment of chemicals. We have used a hierarchical approach to develop class-specific quantitative structure-activity relationship (QSAR) models for the prediction of mutagenicity of a set of 95 aromatic and heteroaromatic amines. The hierarchical approach begins with the simplest molecular descriptors, the topostructural, which encode limited chemical information. The complexity is then increased, adding topochemical, geometric, and finally quantum chemical parameters. We have also added log P to the set of independent variables. The results indicate that the topological parameters, i.e., the topostructural and topochemical indices, explain the majority of the variance, and that the inclusion of log P, geometric, and quantum chemical parameters does not result in significantly improved predictive models.