论文部分内容阅读
本研究初步评价Toxtree、T.E.S.T.、TOPKAT和Derek等QSAR平台,对中草药重要成分实验数据子库的预测可靠性。对数值型数据评价计算外部验证的相关系数Q~2,对分类型数据计算Cooper统计量。实验数据子库中,共83种成分有大鼠经口LD_(50)实验数据,计算T.E.S.T.的Q~2=0.76,TOPKAT的Q~2=0.39。共28种成分有致癌性实验数据,其中致癌性阳性16种,致癌性阴性10种,致癌性不确定2种,实验阳性物比例=16/28=57.1%。Toxtree预测准确性为57.1%,TOPKAT预测准确性为53.6%,Derek预测准确性为50.0%;共46种成分有Ames试验数据,试验阳性14种,阴性32种,实验阳性物比例14/46=30.4%,Toxtree预测准准确性为67.4%,T.E.S.T.预测准确性为76%,TOPKAT预测准确性为71.1%,Derek预测准确性为73.9%。由于不同的原因,对发育毒性、骨髓微核试验、及其他终点难以进行评价。本文提出(Q)SAR预测模型应按OECD验证5项原则进行全面的和标准化的评价。强调要定义明确毒性终点和模型应用域。毒性是复杂的生物现象,不要满足于过于简单的数值。而且,生物毒性数据均具有不确定性。
In this study, the prediction reliability of the experimental data sub-bank of important components of Chinese herbal medicine was preliminary evaluated by QSAR platform such as Toxtree, T.E.S.T., TOPKAT and Derek. Evaluate the numeric data for the correlation coefficient Q ~ 2 for external validation and calculate Cooper statistics for the classification data. In the experimental data sub-database, a total of 83 components of rat oral LD_ (50) experimental data calculated T.E.S.T. Q ~ 2 = 0.76, TOPKAT Q ~ 2 = 0.39. A total of 28 components of carcinogenicity experimental data, including 16 carcinogenic positive, carcinogenic negative 10, carcinogenicity indeterminate 2, the proportion of experimental positive = 16/28 = 57.1%. Toxtree prediction accuracy was 57.1%, TOPKAT prediction accuracy was 53.6%, Derek prediction accuracy was 50.0%; a total of 46 components were Ames test data, 14 positive tests, 32 negative tests, the ratio of experimental positive 14/46 = 30.4%, Toxtree prediction accuracy was 67.4%, TEST prediction accuracy was 76%, TOPKAT prediction accuracy was 71.1%, Derek prediction accuracy was 73.9%. Due to different reasons, it is difficult to evaluate developmental toxicity, bone marrow micronucleus test, and other end points. This paper proposes that the (Q) SAR prediction model should be comprehensive and standardized according to the five principles of OECD verification. Emphasis is placed on defining clear end points of toxicity and model application domains. Toxicity is a complex biological phenomenon, do not be content with overly simple values. Moreover, the biotoxicity data are both uncertain.