摘要
目的探讨神经网络方法应用于5,7,8位取代的喹诺酮类化合物定量构效关系研究的效果。方法利用Matlab软件包构建一个三层的BP神经网络,对数据集进行计算,并将结果与线性回归进行比较。结果神经网络法的误差平方和为0.3042, 小于线性回归法,预测的相关系数为0.86。结论神经网络方法获得了比回归模型更精密的拟合。
Objective To explore the effect of neural network( NN) on the study of structure - activity relationship of 5,7,8 - substituted quinolones .Methods A three- layer BP neural network was built by the Matlab software.The data were calculated and analysed.The results of present study were compared with those of the study by linear regression. Results NN'SSE (sum of square errors) was 0.3042, less than that of linear regression. The predicted correlation coefficient was 0.86. Conclusion The fitted performance of NN was better than that of linear regression.
出处
《华南国防医学杂志》
CAS
2005年第1期48-51,共4页
Military Medical Journal of South China
基金
国家自然科学基金资助项目(60371034)
