SVR-CKNN预测用于HIV-1蛋白酶抑制剂QSAR建模

SVR-CKNN Applied in QSAR Model of HIV-1 Protease Inhibitors

为提高药物定量构效关系(QSAR)模型预测精度,发展了一种新的QSAR建模方法SVR-CKNN。该法基于支持向量机回归(SVR)自动筛选化合物结构描述符,以k-最近邻建立多个子模型实施组合预测(CKNN)。应用于49种H IV-1蛋白酶抑制剂QSAR研究,留一法预测结果表明SVR-CKNN预测精度明显优于多元线性回归(MLR)、逐步回归(SLR)、偏最小二乘回归(PLS)和神经网络(BP-ANN)等传统模型。SVR-CKNN基于结构风险最小,具非线性、适于小样本、泛化推广能力强、稳定性好、不依赖操作者经验等诸多优点,在药物设计等研究中应用前景广泛。

In order to improve the predication precision of quantitative structure-activity relationship （QSAR） model, a novel combinatorial k-nearest neighbor method based on support vector machine regression （SVR-CKNN） was proposed, which could screen descriptors automatically and then builds several k-nearest neighbor models for combinatorial forecast. This model applied to data of 49 HIV-1 protease inhibitors, and it was then tested by jackknife method, The result showed new model was more precise than traditional, as it had the advantages in strong generalization ability, fitting small simples and independent on user＇ s experience. The novel combination model, therefore, are broad prospect of application in QSAR for drug design.