The occurrence of contaminants in natural waters is a potential threat to the environment. Since contaminants are commonly present as mixtures, numerous interactions may occur, resulting in lower or, more dangerously, higher toxicity, by comparison with single substances. The toxicity of multicomponent systems can be determined experimentally, but toxicity prediction by suitable models is faster, environmentally friendly and less expensive. Here we review approaches and models, which can be utilized in assessing toxicity of chemical mixtures. In the first part, the assessment of toxicity of chemical mixtures and possible interactions between mixture constituents are discussed. The second part covers conventional modeling, including the simplest, and most common toxicity models, namely concentration addition and independent action models, and derived integrated models. The third part presents advanced toxicity modeling. We review the quantitative structure–activity relationship (QSAR) approach and its elements: calculation of molecular descriptors and their selection with principal component analysis and genetic algorithm. Modeling with artificial neural networks is also discussed. We present hybrid models which combine the fuzzy set theory approach with the conventional concentration addition and independent action models. We conclude that conventional models: concentration addition and independent action model, are still most commonly used; integrated models are more accurate compared to conventional ones, even though their application requires more data; advanced numerical methods such as genetic algorithm, neural networks, and fuzzy set theory give a new perspective on toxicity prediction, and no universal tool for toxicity assessment has been developed so far.

天然水中污染物的出现是对环境的潜在威胁。由于污染物通常以混合物形式存在，因此与单一物质相比，可能会发生多种相互作用，从而导致更低或更危险的更高毒性。可以通过实验确定多组分系统的毒性，但是通过合适的模型进行的毒性预测更快，对环境友好并且更便宜。在这里，我们回顾了可用于评估化学混合物毒性的方法和模型。在第一部分中，讨论了化学混合物的毒性评估以及混合物成分之间可能的相互作用。第二部分介绍了常规建模，包括最简单，最常见的毒性模型，即浓度增加和独立作用模型，以及派生的集成模型。第三部分介绍了高级毒性建模。我们回顾了定量构效关系（QSAR）方法及其要素：分子描述符的计算以及主成分分析和遗传算法的选择。还讨论了用人工神经网络建模。我们提出了混合模型，该模型将模糊集理论方法与常规浓度添加和独立动作模型相结合。我们得出的结论是，常规模型：浓度增加和独立作用模型仍然是最常用的。尽管集成模型的应用需要更多数据，但与传统模型相比，它们更准确；先进的数值方法，例如遗传算法，神经网络，