COX2抑制剂活性预测中的拟合方法比较

Comparision of fitting methods in prediction of COX2 inhibitors' effect

在COX2抑制剂活性预测问题中,传统的多元线性回归方法对非线性拟合问题存在较大局限性,以98种三环系COX2抑制剂分子集合中的数据分析预测为例,考察反向传播(BP)网络与支持向量回归(SVR)方法的性能,证明BP网络与SVR方法在COX2抑制剂活性预测方面的性能均优于广泛采用的多元线性回归方法,具有良好的应用潜力。而且SVR方法可以克服BP网络过拟合的问题,更适于小样本高参量的问题。

In the research of COX2 inhibitors effect prediction, there is obvious limitation of the multiple response method applied for the fitting process. Applied back-propagation （BP） neural networks and supporting-vector regression （SVR） method to the effect prediction on the set of 98 3-circle-series COX2 inhibitors in order to check the two methods＇ performance on the fitting of non-linear function. The two methods＇ performance are proved that they are better than the multiple response method and hava good application prospect. And the SVR method can overcome the over-fit problem so as it is more suitable for the problem with few samples and many variables.