化合物ADMET性质预测平台的构建  

Development of the platform for prediction of chemical ADMET properties

作　　者：李晓 李达 周雪松 赵勇 

机构地区：[1]北京市科学技术研究院北京市计算中心,北京100094 [2]北京北科德源生物医药科技有限公司,北京100094

出　　处：《生物信息学》2017年第3期179-185,共7页Chinese Journal of Bioinformatics

基　　金：北京市科技型中小企业促进专项项目(Z16010101204)

摘　　要：在药物研发早期阶段对化合物成药性和安全性进行评估,对于提高药物研发成功率、降低研发成本具有十分重要的意义。为了能够帮助药物研究工作者快速准确地判断候选化合物的成药性与安全性,开发了一个基于计算机方法的化合物ADMET性质预测平台。首先,通过文本挖掘的方法收集了化合物药代动力学性质和毒性(ADMET)的高质量实验数据。然后,根据原始文献复原了13个预测模型,同时采用支持向量机方法自建了15个具有较高预测能力的计算模型。最后,基于分布式架构,结合高性能计算集群优势,开发了化合物ADMET性质预测平台(http://www.vslead.com/?r=admet/index),用于预测28种重要的化合物ADMET性质。研究者可以使用这一平台快捷方便地对药物研究中比较重要的ADMET性质进行预测,在药物研发早期对候选化合物进行成药性评价和风险评估,有助于提高药物研的成功率,节省研发时间和经费的投入。It is very essential to estimate the druggability and toxicity of chemicals in the early stages of drug discovery. In order to make it available for drug researchers to estimate the druggability and toxicity of candidate compounds rapidly and accurately, we developed a platform for estimation of chemical ADMET properties based on in silico methods in this study. Firstly, we collected quality data of chemical pharmacokinetic and toxic properties （ADMET） with text mining approaches. Then, 13 predictive models were reproduced on the basis of original literature. Meanwhile, another 15 models were built with support vector machine （SVM）. Finally, we developed a platform for estimation of chemical ADMET properties based on distributed storage architectures. This tool could be used to predict 28 important ADMET related properties quickly and easily, and its application in drug discovery could be helpful to reduce the time and money.

关 键 词：药物研发 ADMET性质 机器学习 

分 类 号：R91[医药卫生—药学]