基于单类图分类的化合物分类方法

摘要

To reduce the cost of screening compounds and structure optimization in the design process of new drugs, and improve the accuracy of screening and optimization, one-class graph classification method for compound classification was proposed.Considering the strict topology structure of the compound, the method of optimization of frequent subgraph was used to select the feature subgraph, and the closed frequent subgraph which was more complete at each stage of support was adaptively mined, and the extracted frequent subgraphs were inputted as the feature samples.The classification algorithm of Adaboost was used to train one-class support vector machine classification model.Experimental results show that the proposed method can extract frequent substructures with high correlation and significantly reduce the feature space after frequent subgraph mining, and effectively improve the accuracy and generalization of classification.%为减少新药设计过程中化合物筛选与结构优化的成本, 提高筛选与优化的精确性, 提出一种用于化合物分类的单类图分类方法.对化合物的严格拓扑结构提出优化频繁子图的方法选取特征子图, 自适应地挖掘每阶段支持度下信息量表示更全的闭频繁子图, 将提取的频繁子图作为特征样本输入, 通过Adaboost集成一类支持向量机分类算法训练分类模型.实验结果表明, 该方法可以明确提取相关性较高的频繁子结构, 显著降低频繁子图挖掘后的特征空间, 有效提高分类的准确性和泛化性.