Schüürmann Gerrit, Aptula Aynur O, Kühne Ralph, Ebert Ralf-Uwe
Department of Chemical Ecotoxicology, UFZ Centre for Environmental Research, Permoserstrasse 15, 04318 Leipzig, Germany.
Chem Res Toxicol. 2003 Aug;16(8):974-87. doi: 10.1021/tx0340504.
For a set of 220 phenols with literature data on their toxicity and associated mode of action (MOA) toward the ciliate Tetrahymena pyriformis, a stepwise classification scheme was developed that allows the identification of four MOAs from molecular hydrophobicity and AM1-based quantum chemical descriptors, employing linear discriminant analysis or binary logistic regression. Taking the AM1 lowest unoccupied molecular orbital energy as the only parameter, an initial separation of polar narcotics and proelectrophiles from oxidative uncouplers and soft electrophiles is correct to 97%, and for the subsequent discrimination between polar narcotics and proelectrophiles as well as between oxidative uncouplers and soft electrophiles, 99 and 98% correct classifications are achieved using three and two molecular descriptors, respectively. The results are discussed in terms of detailed contingency table statistics and with respect to relationships between molecular descriptors and mechanisms of toxicity. Statistical model evaluation includes simulated external validation employing complementary subset models.
对于一组220种具有关于其对梨形四膜虫毒性及相关作用模式（MOA）的文献数据的酚类化合物，开发了一种逐步分类方案，该方案利用线性判别分析或二元逻辑回归，能够从分子疏水性和基于AM1的量子化学描述符中识别出四种作用模式。仅以AM1最低未占分子轨道能量作为参数，将极性麻醉剂和亲电试剂与氧化解偶联剂和软亲电试剂初步分离的正确率为97%，对于随后极性麻醉剂和亲电试剂之间以及氧化解偶联剂和软亲电试剂之间的判别，分别使用三个和两个分子描述符时，正确分类率分别达到99%和98%。根据详细的列联表统计以及分子描述符与毒性机制之间的关系对结果进行了讨论。统计模型评估包括使用互补子集模型进行模拟外部验证。
