摘要
用人工神经网络中误差反向传播网络(BPNN)和径向基函数网络(RBFNN)对甲基、烷基、环戊并及环己并及胆蒽系化合物的致癌性强弱进行了分类.采用的输入参数为单个原子能(IAE)、电子能(EE)、生成热(HOF)、原子最高正电荷(QMAX)、原子最低负电荷(QMIN)、最高占有轨道能量(HOMO)、最低未占有轨道能量(LUMO)、偶极矩(DIP)、水合能(HE)、疏水性参数(logP)、分子表面积(SA)、极化率(Polar)、代谢活性区中心碳原子离域能(?E1)、亲电活性区中心碳原子离域能(?E2)和分子中脱毒区总数(n).BP网络采用tan-sigmoid函数;;RBF网络采用Quadratic和InverseQuadratic函数.两种模型的分类准确率均达80%以上.
Arti?cial neural networks with back-propagation learning algorithm and radial basis function neural networks were applied for classifying the carcinogenicity of polycyclic aromatic hydrocarbons and cholanthrene. In this paper, 15 parameters were used as input factors of neural networks. These parame- ters are: LAE, EE, HOF, QMAX, QMIN, HOMO, LUMO, DIP, HE, logP, SA, Polar, ?E1, ?E2 and n. In BP networks, the tan-sigmoid function was used as the transfer function, and the Quadratic and Inverse Quadratic function were used as the transfer functions of RBFNN. The accuracy of classi?cation by all model were more than 80 percent. All the results indicated that the proposed models were suitable to classify the carcinogenicity of polycyclic aromatic hydrocarbons and cholanthrene.
出处
《兰州大学学报（自然科学版）》
CAS
CSCD
北大核心
2004年第1期38-44,共7页
Journal of Lanzhou University(Natural Sciences)
基金
国家自然科学基金资助项目(20275014).
