‘Ideal correlations’ for the predictive toxicity to Tetrahymena pyriformis

Predictive models for toxicity to Tetrahymena pyriformis are an important component of natural sciences. The present study aims to build up a predictive model for the endpoint using the so-called index of ideality of correlation (IIC). Besides, the comparison of the predictive potential of these models with the predictive potential of models suggested in the literature is the task of the present study. The Monte Carlo technique is a tool to build up the predictive model applied in this study. The molecular structure is represented via a simplified molecular input-line entry system (SMILES). The IIC is a statistical characteristic sensitive to both the correlation coefficient and mean absolute error. Applying of the IIC to build up quantitative structure–activity relationships (QSARs) for the toxicity to Tetrahymena pyriformis improves the predictive potential of those models for random splits into the training set and the validation set. The calculation was carried out with CORAL software (http://www.insilico.eu/coral). The statistical quality of the suggested models is incredibly good for the external validation set, but the statistical quality of the models for the training set is modest. This is the paradox of ideal correlation, which is obtained with applying the IIC. The Monte Carlo technique is a convenient and reliable way to build up a predictive model for toxicity to Tetrahymena pyriformis. The IIC is a useful statistical criterion for building up predictive models as well as for the assessment of their statistical quality.针对梨形四膜虫（Tetrahymena pyriformis）毒性的预测模型是自然科学领域的重要组成部分。本研究旨在利用相关性理想指数（index of ideality of correlation, IIC）构建针对该受试终点的预测模型。此外，对比本研究所建模型与文献已报道模型的预测性能亦是本研究的任务之一。本研究采用蒙特卡洛（Monte Carlo）技术构建预测模型，分子结构通过简化分子线性输入系统（simplified molecular input-line entry system, SMILES）进行表征。相关性理想指数（IIC）是一种同时对相关系数与平均绝对误差均敏感的统计特征。将IIC应用于构建针对梨形四膜虫毒性的定量构效关系（quantitative structure–activity relationships, QSAR）模型时，可提升模型在随机划分为训练集与验证集场景下的预测性能。所有计算均通过CORAL软件（http://www.insilico.eu/coral）完成。所提模型在外部验证集上的统计性能极佳，但在训练集上的统计表现则较为一般，这便是应用IIC所得到的理想相关性悖论。蒙特卡洛技术是构建梨形四膜虫毒性预测模型的便捷且可靠的方法。相关性理想指数（IIC）作为构建预测模型以及评估其统计性能的有效统计准则，具备良好的实用价值。