用自适应偏最小二乘回归为药物定量构效关系建模

To Model the Quantitative Structure-Activity Relationship of Drug with Adapting Partial Least Square Regression

提出了一种自适应偏最小二乘回归 (APLSR)方法。对于指定的预测对象 ,APLSR将根据样本在结构参数空间中的分布 ,分析它们对预测对象的预报能力 ,自适应地为各个样本分配权值 ,并从样本数据中提取和选用PLS成分 ,从而实施自适应加权PLSR ,以获得预报性能良好的模型。作者将APLSR应用于含硫苯衍生物的QSAR建模 ,取得了令人满意的效果。

There usually exist the nonlinear quantitative structure-activity relationships ( QSAR) of drug and significant correlation among structure parameters of the drug. Sometimes, the multicollinearity is even formed among the structure parameters. Thus, the ideal regression model of QSAR with high predicting correctness can't be obtained by the method of global linear regression. At the same time, the benefits of QSAR models are valued by their predicting abilities. An adapting partial least square regression (APLSR) was proposed to model the drug's QSAR. In order to obtain the QSAR model with high predicting correctness for some predicting sample, the different predicting contribution ratio of modeling samples for the predicting sample was taken into account as well as the number of latent variables by APLSR. When APLSR was employed for the predicting sample, each modeling sample was weighted according to its different ratio of predicting contribution for the predicting sample and the optimal number of the latent variables was obtained according to the predicting ability of model. Finally, a typical example of modeling the QSAR of substituted aromatic sulfur derivatives was employed to verify the effectiveness of APLSR. The satisfactory result was obtained.