用支持向量机回归(SVR)方法研究了48个黄酮类化合物醛糖还原酶抑制活性的定量构效关系。建模过程中利用留一法交叉验证(LOOCV)优化了核函数的类型、惩罚系数C和不敏感函数ε.所建模型最终采用了227个变量中的7个:dChivps9, ESHaaCH,EsssCH2,n2pag[1,2],degree2,I’3和I’4。所得SVR模型的预测相对误差为0.0622,小于多元线性回归(MLR)和偏最小二乘法(PLS),以及文献报道模型的预测相对误差。 The quantitative structure-activity relationship of aldose reductase inhibitory activity of 48 flavonoids was studied by support vector machine regression (SVR). In the modeling process, the type of kernel function, the penalty coefficient C and the insensitive function ε are optimized by leaving one method of cross-validation (LOOCV). The model finally adopts 7 of 227 variables: dChivps9, ESHaaCH, EsssCH2, n2pag [1,2], degree2, I’3 and I’4. The predicted relative error of the obtained SVR model is 0.0622, which is smaller than the multiple linear regression (MLR) and partial least squares (PLS), and the prediction relative error of the reported model.