基于局部核函数与全局核函数支持向量回归优化小样本QSAR建模

Improve performance of SVR model on small sample QSAR study with local kernel and global kernel

为提高小样本定量构效关系(QSAR)预测精度,基于支持向量机全局核函数与局部核函数提出了一种新的建模方法:先依不同核函数筛选描述符,再依保留描述符构建支持向量机回归(SVR)子模型.子模型预测活性值与实验值组成混合样本.以均方误差(MSE)最小为原则,对混合样本再次基于SVR实施核函数寻优与子模型筛选,基于最优核函数和保留子模型以留一法完成预测.对2个小样本体系的QSAR研究表明,该方法兼具局部核函数和全局核函数的优点,既有较强的学习能力,又有较好的推广能力,预测精度高,稳定性好.

To improve the performance of QSAR model on small sample sets, a novel combinatorial model based on global kernel and local kernel was proposed in this paper. Firstly, descriptors were screened according to different kernel function. And then, SVR sub-model construction based on retained descriptors. The prediction results of sub- model and the experiment results assembled to be a mixed sample set. Kernels and descriptors optimization based on SVR was evaluated by the rule of minimum MSE value. The forecast was carried out through the leave-one-out method based on optimized kernel and sub-model. The results of QSAR modeling on two small sample sets showed that the combinatorial model has precise and stable predicting ability.