摘要
采用原子对空间距离法表征化合物的三维结构信息.提出了计算端点的频数的方法,与段的频数相比,该方法能更好地描述原子对频数.以不同段的距离,分别采用上述两种频数计算得到化合物的相似度矩阵,然后把相似度衍生为新的变量.运用多元回归分析和人工神经网络分别构造了预测数学模型,并对所得到的预测结果进行了比较.这两种频数均较好地预测了HEPT类化合物的活性.
In this article, atom-pairs of compounds, which include abundant three-dimensional information of molecules, were calculated. Vertex＇s frequency of atom-pairs space distance was applied to describe the frequency of atom pairs, which is better than segment＇s frequency of atom-pairs. Molecular similarity matrixes based on the two frequencies of atom-pairs in different distances of segments were calculated, respectively, and then these similarities were taken as the new variables. The mathematical models were built by using multiple regression analysis and artificial neural networks and the results were compared. The results of predictions of the activities of HEPT derivatives in both two frequencies are satisfactory.
出处
《高等学校化学学报》
SCIE
EI
CAS
CSCD
北大核心
2008年第7期1438-1442,共5页
Chemical Journal of Chinese Universities
基金
国家自然科学基金(批准号:20375039)资助
关键词
QSAR
原子对空间距离法
分子相似度
多元回归分析
人工神经网络
QSAR
Atom-pairs space distance method
Molecular similarity
Multiple regression analysis
Artificial neural network
