支持向量机分类和回归用于肽的QSAR研究

Classification and Regression with Support Vector Machine as Applied to QSARs of Peptides

使用支持向量机技术对两类肽化合物体系进行了分类和回归研究,并将其系统地与K最邻近法、多元线性回归、偏最小二乘、人工神经网络进行了比较。结果表明,对于小样本、非线性问题,支持向量机具有较强的稳定性能及泛化能力,在大多数情况下能够得到优于传统方法的建模效果。对于分类问题,支持向量机对训练集和测试集都达到了100%的分类正确率;对于回归问题,支持向量机虽对训练集样本拟合效果略低于人工神经网络,但对外部测试集却表现出较强的预测能力。

Support vector machine （SVM） is employed to classification and regression for analyzing two various kinds of peptide analogues. Simultaneously, comparisons are systematically done on the results obtained with different methods of K-nearest neighbor method （KNN）, multiple linear regression （MLR）, partial least square regression （PLS） and artificial neural network （ANN）; it is suggested that SVM possesses modeling stability and generalization ability, especially when applied to investigating both nonlinear questions and small sampling, yielding superior modeling results. For classification, SVM has achieved a 100% resolution for both the training set and the testing set; while for regression, SVM has quite stronger predict abilities for samples in the external prediction set although it slightly weaker than ANN for internal calibration set, respectively.