Ren Shijin
Center for Environmental Biotechnology, 676 Dabney Hall, University of Tennessee, Knoxville, TN 37996-1605, USA.
Toxicol Lett. 2003 Oct 15;144(3):313-23. doi: 10.1016/s0378-4274(03)00236-4.
In this study, the use of decision tree in classifying and predicting the aquatic toxicity mechanisms of phenols was investigated. Four mechanisms including polar narcosis, respiratory uncoupling, pro-electrophilicity, and soft electrophilicity were involved. Using molecular descriptors as splitting variables, a three level decision tree with six terminal nodes was obtained. The tree model first separated polar narcosis/pro-electrophilicity from respiratory uncoupling/soft electrophilicity by E(lumo) in the first level of the tree. In subsequent levels of the tree, polar narcosis was separated from pro-electrophilicity by N(hdon) and E(homo), and respiratory uncoupling was separated from soft electrophilicity by E(lumo) and logK(ow). Validation of the decision tree approach indicated that the overall mechanism prediction accuracy was approximately 85%. The decision tree model had the advantage of easy interpretation.
在本研究中，对决策树在酚类水生毒性机制分类和预测中的应用进行了研究。涉及极性麻醉、呼吸解偶联、亲电前体性和软亲电性四种机制。以分子描述符作为分裂变量，得到了一个具有六个终端节点的三级决策树。该树模型首先在树的第一级通过最低未占轨道能量（E(lumo)）将极性麻醉/亲电前体性与呼吸解偶联/软亲电性分开。在树的后续级别中，通过供电子数（N(hdon)）和最高已占轨道能量（E(homo)）将极性麻醉与亲电前体性分开，通过最低未占轨道能量（E(lumo)）和辛醇-水分配系数对数值（logK(ow)）将呼吸解偶联与软亲电性分开。决策树方法的验证表明，总体机制预测准确率约为85%。决策树模型具有易于解释的优点。
