有机化合物吸着系数(Koc)估算模型比较

Comparison of KOC prediction models on the based of fragment Constants and molecular connectivity indices

基于 592种化合物的Koc实测数据 ,分别采用片段常数法和分子连接性指数法建立了其估算模型 .根据模型残差分析了两种模型的预测误差 ,并运用 2种方式的jackknife检验比较了它们的稳健性 .片段常数模型涉及较多变量 ,其应用受到有限的已知片段常数以及烦琐的片段划分过程的限制 ,相比之下 ,分子连接性指数模型变量数较少 ,其应用时只需了解化合物的结构式 .片段常数模型的估算精度略高于分子连接性指数模型 ,两者的差别主要体现在logKoc<1和logKoc >4的化合物上 .分子连接性指数模型的稳健性比片段常数模型高 .在依次删除一类特定类别化合物时 ,其估算误差的变异幅度均小于片段常数模型 .

Two modeling approaches, namely fragment constant (FC) and molecular connectivity indices (MCI), were compared for their accuracy and robustness in predicting soil adsorption coefficient ( Koc ). The criteria of the comparison were regression residuals. Jackknife test was thus applied for the robustness evaluation. The results showed that the FC model was better than the MCI one in terms of estimation accuracy, especially for the chemicals with log Koc smaller than 1 and greater than 4. On the other hand, the MCI model was more robust than the FC one as far as the sensitivity of output to input is considered.