Kleandrova Valeria V, Luan Feng, Speck-Planche Alejandro, Cordeiro M Natália D S
REQUIMTE/Department of Chemistry and Biochemistry, University of Porto, 4169-007 Porto, Portugal.
Mini Rev Med Chem. 2015;15(8):677-86. doi: 10.2174/1389557515666150219143604.
The assessment of acute toxicity is one of the most important stages to ensure the safety of chemicals with potential applications in pharmaceutical sciences, biomedical research, or any other industrial branch. A huge and indiscriminate number of toxicity assays have been carried out on laboratory animals. In this sense, computational approaches involving models based on quantitative-structure activity/toxicity relationships (QSAR/QSTR) can help to rationalize time and financial costs. Here, we discuss the most significant advances in the last 6 years focused on the use of QSAR/QSTR models to predict acute toxicity of drugs/chemicals in laboratory animals, employing large and heterogeneous datasets. The advantages and drawbacks of the different QSAR/QSTR models are analyzed. As a contribution to the field, we introduce the first multitasking (mtk) QSTR model for simultaneous prediction of acute toxicity of compounds by considering different routes of administration, diverse breeds of laboratory animals, and the reliability of the experimental conditions. The mtk-QSTR model was based on artificial neural networks (ANN), allowing the classification of compounds as toxic or non-toxic. This model correctly classified more than 94% of the 1646 cases present in the whole dataset, and its applicability was demonstrated by performing predictions of different chemicals such as drugs, dietary supplements, and molecules which could serve as nanocarriers for drug delivery. The predictions given by the mtk-QSTR model are in very good agreement with the experimental results.
急性毒性评估是确保在药物科学、生物医学研究或任何其他工业领域有潜在应用的化学品安全性的最重要阶段之一。在实验动物身上已经进行了大量且无差别的毒性试验。从这个意义上说，涉及基于定量结构活性/毒性关系（QSAR/QSTR）模型的计算方法有助于合理安排时间和资金成本。在此，我们讨论过去6年中在使用QSAR/QSTR模型预测实验动物中药物/化学品急性毒性方面取得的最重要进展，这些模型采用了大量且异质的数据集。分析了不同QSAR/QSTR模型的优缺点。作为对该领域的贡献，我们引入了首个多任务（mtk）QSTR模型，通过考虑不同给药途径、多种实验动物品种以及实验条件的可靠性来同时预测化合物的急性毒性。mtk - QSTR模型基于人工神经网络（ANN），能够将化合物分类为有毒或无毒。该模型正确分类了整个数据集中1646个案例中的94%以上，并且通过对不同化学品（如药物、膳食补充剂以及可作为药物递送纳米载体的分子）进行预测证明了其适用性。mtk - QSTR模型给出的预测结果与实验结果非常吻合。
