Shoji Ryo, Kawakami Masato
Department of Chemical Science and Engineering, Tokyo National College of Technology, 1220-2 Kunugida-machi, Hachioji, Tokyo, 193-0997, Japan.
Mol Divers. 2006 May;10(2):101-8. doi: 10.1007/s11030-005-9005-1. Epub 2006 Jun 27.
In order to evaluate human carcinogenic risks, genotoxicity data such as animal cancer bioassay are often not available. In this study, to assess the relevance of indicator of carcinogenic risks, we used the "molecular diversity approach" to estimate the genotoxicity based upon Salmonella genotoxicity test using the umu test and systemic toxicity data of the 82 environmental chemicals predicted by neural network simulation. The 82 environmental chemicals were randomly selected for this study according to the production and usage in Japan. Even in this challenging trial for QSTR (Quantitative Structure Toxicity Relationship) study, approaches using artificial neural networks can account for about 94% of the variation in the genotoxicity results derived by the umu-test.
为了评估人类致癌风险，通常无法获得动物癌症生物测定等遗传毒性数据。在本研究中，为了评估致癌风险指标的相关性，我们使用“分子多样性方法”，基于沙门氏菌遗传毒性试验（umu试验）以及神经网络模拟预测的82种环境化学品的全身毒性数据来估计遗传毒性。根据日本的生产和使用情况，随机选择了这82种环境化学品用于本研究。即使在这项对定量结构-毒性关系（QSTR）研究具有挑战性的试验中，使用人工神经网络的方法也能解释umu试验得出的遗传毒性结果中约94%的变化。
