支持向量机用于多氯代萘毒性的定量构效研究

Quantitative Structure and Activity Relationship Studies on the Toxicity of Polychlorinated Naphthalenes Using Support Vector Machines

用偏最小二乘法(PLS)和留一交叉验证从90多个量子化学参数中筛选出极化率、分子量、部分原子上的净电荷、静电势等作为描述符,应用支持向量机(SVM)对20个多氯代萘同系物的三组毒性数据分别建立了定量构效关系模型。所得模型的交叉验证相关系数的平方分别为0.805、0.890、0.936。并将偏最小二乘法建模所得结果与之进行比较,结果表明,SVM预报能力优于PLS。

By using partial least square （PLS） with leave-one-out cross-validation, several descriptors, such as polarizability,molecular weight, net atomic charge and electrostatic potential on some atoms and so on, were selected from more than 90 quantum chemical descriptors. Using support vector machine （SVM）, quantitative structure ＆ activity relationship （QSAR） models were established for three sets of the dioxin-like toxicities （relative potencies） of 20 polychlorinated naphthalenes （PCNs）. The squared cross-validation correlation coefficients （q^2） of the three models were 0. 805,0.890 and 0. 936,respectively. PLS has been utilized to compare with the results obtained by SVM. Therefore, the result indicated that SVM were better than PLS.