支持向量机用于芳烃类化合物对芳烃受体亲和性QSAR研究

尝试将支持向量机(SVM)应用于3种典型芳烃类环境毒物(PCDD,PCDF和PCB)定量构效关系研究,通过对芳烃受体亲和性考察,结果发现该组样本的生物活性在一定程度上与分子电性距离矢量具有非线性联系.SVM对内部和外部样本都具良好稳定性能和预测能力:所得模型拟合、交叉检验、外部预测复相关系数及均方根误差分别为R2cum=0.922、Q2cum=0.825、Q2ext=0.834和RMSext=0.531将其与文献报道及多元线性回归、偏最小二乘、人工神经网络进行比较,结果表明对小样本、非线性问题SVM具较强拓展性及泛化能力,故在环境毒物评价和控制中具有广阔应用前景.

QSAR study on applying support vector machine to binding affinity of Ah receptor with aromatic compounds

Support vector machine (SVM) was employed to investigate quantitative structure-activity relationship (QSAR) of three typical kinds of aromatic compounds (PCDDs, PCDFs and PCBs). It was found that binding affinity of Ah receptor with the aromatic compounds was nonlinearly relative to molecular electronegativity distance vector (MEDV) at a given conditions. The established model by SVM was proved to be of stability and predictability since its correlation coefficients of fitting (R2_ cum), cross validation (Q2_ cum) and external prediction (Q2_ ext) and error root mean square errors (RMS_ ext) were 0.922, 0.825, 0.834 and 0.531 respectively. Therefore, it was proved to do good to induct especially for nonlinear and small sampling questions through comparison among reference reports and different methods such as multiple linear regression (MLR), partial least square regression (PLS) and artificial neural network (ANN). Therefore, SVM is expected to have broad prospect in evaluating and controlling environmentally toxic compounds.