论文部分内容阅读
首先对同系有机化合物进行定量结构毒性相关研究,选取263个芳香类化合物的生物毒性数据作为建模样本,分别采用有监督的模式识别方法—K最近邻方法和无监督的模式识别方法—K均值聚类方法对相关样本数据进行模式识别分析,对模式识别后的每一类样本分别运用多元线性回归、偏最小二乘和人工神经网络方法进行QSTR预测模型的建立。然后又对非同系化合物进行QSTR研究,选取90个有机分子的肝癌致毒性数据作为建模样本,仍采用上述方法进行建模研究。在上述研究工作的基础之上,得出了对同系及非同系有机物建立QSTR模型的有关规律。
Firstly, the quantitative structural toxicity of homologous organic compounds was studied. The bio-toxicity data of 263 aromatic compounds were selected as the modeling samples. The supervised model recognition method-K nearest neighbor method and the unsupervised pattern recognition method were used respectively. K-means The clustering method performs pattern recognition analysis on the related sample data, and establishes the QSTR prediction model by using multiple linear regression, partial least squares and artificial neural network respectively for each type of pattern recognition. Then the non-homologous compounds QSTR research, selection of 90 organic molecules of liver cancer toxicity data as a model sample, still using the above method modeling. Based on the above research work, the related rules of establishing QSTR model for homologous and nonhomologous organic compounds are obtained.