Ren Shijin, Schultz T Wayne
Department of Chemical Engineering, 419 Dougherty Engineering Building, University of Tennessee, Knoxville, TN 37996-2200, USA.
Toxicol Lett. 2002 Mar 24;129(1-2):151-60. doi: 10.1016/s0378-4274(01)00550-1.
The most successful quantitative structure-activity relationships (QSARs) have been developed by separating toxicants by their mechanisms of action (MOAs). However, since the activity of a chemical compound on an organism is dependent upon several physical, chemical and biological factors, among which interactions may also exist, the MOA of a compound is not easily determined. In this study, the use of discriminant analysis and logistic regression in distinguishing between narcotic and reactive compounds was investigated. The discriminating variables included hydrophobicity (log(K(ow))) and electrophilicity descriptors (S(av)(N), E(HOMO), and E(LUMO)). Classification results showed that logistic regression gave a smaller total error rate compared to discriminant analysis. Since the value of the descriptors can be calculated, the classification methods can be used in predictive toxicology.
最成功的定量构效关系（QSARs）是通过根据作用机制（MOAs）对毒物进行分类而建立的。然而，由于化合物对生物体的活性取决于多种物理、化学和生物学因素，其中可能还存在相互作用，因此化合物的作用机制并不容易确定。在本研究中，研究了使用判别分析和逻辑回归来区分麻醉性化合物和反应性化合物。判别变量包括疏水性（log(K(ow))）和亲电性描述符（S(av)(N)、E(HOMO)和E(LUMO)）。分类结果表明，与判别分析相比，逻辑回归的总错误率更小。由于描述符的值可以计算，因此这些分类方法可用于预测毒理学。
