QSPR as a random event: solubility of fullerenes C[60] and C[70

Abstract

Correlations of criteria of predictive potential of models for solubility of fullerenes C[60] and C[70] observed for the calibration (visible) set with the determination coefficient of corresponding models for validation sets (external, invisible). The Index of Ideality of Correlation (IIC) gave the best correlation with the determination coefficient for the validation set. The IIC was involved in the Monte Carlo optimization used to build up one-variable quantitative structure-property relationships (QSPR) to predict the solubility of fullerenes C[60] and C[70]. This considerably improved the predictive potential of models for this solubility. According to principle “QSPR is a random event”, corresponding computational experiments, which are aimed to build up model, were carried out for group of ten random splits into the training and validation sets. These calculations are source of usual data related to assessment of predictive models, such as the statistical quality for the training and validation sets and mechanistic interpretation of models in terms of structural alerts which are promoter of increase (or decrease) for an endpoint.