ABSTRACT

The persistence of organic pollutants is an important environmental property due to the extended possibility to have an impact of corresponding substances. In many cases, the experimental values of the thousands of contaminants are missing. The object of the study is novel computational modelling for air pollutions. Quantitative structure–property relationship (QSPR) for air half-life has been built using the Monte Carlo method with applying the index of ideality of correlation (IIC). The basis of the predictive model of air half-life is the representation of the molecular structure by simplifying molecular input-line entry system (SMILES) and numerical data on the above endpoint (expressed by hours) converted to a decimal logarithm. The statistical quality of the model has been checked up with different validation metrics and is quite good. Paradoxically, the improvement of the statistical quality via the IIC for the validation set is done in detriment to the training set. The new model has performed better than those obtained previously on the same set of compounds, for the prediction of new compounds in the validation set. Some semi-quantitative indicators for the mechanistic interpretation of the model are suggested.

摘要

有机污染物的持久性是一项重要的环境特性，因为其影响相应物质的可能性很大。在许多情况下，数以千计的污染物的实验值是缺失的。该研究的目的是针对空气污染的新型计算模型。空气半衰期的定量结构-性质关系 (QSPR) 已使用蒙特卡罗方法和应用理想相关指数 ( IIC）。空气半衰期预测模型的基础是通过简化分子输入线输入系统 (SMILES) 和将上述端点上的数值数据（以小时表示）转换为十进制对数来表示分子结构。该模型的统计质量已经通过不同的验证指标进行了检查，并且非常好。矛盾的是，通过IIC对验证集的统计质量的改进不利于训练集。在预测验证集中的新化合物时，新模型的性能优于之前在同一组化合物上获得的模型。提出了一些用于模型机械解释的半定量指标。