中国药物警戒 ›› 2024, Vol. 21 ›› Issue (7): 776-780.
DOI: 10.19803/j.1672-8629.20230409

• 基础与临床研究 • 上一篇    下一篇

基于规定日服用剂量改进药物肝毒性预测方法的研究

胡笑文, 张才煜, 王峰峰, 濮恒婷, 刘阳#, 陈华*   

  1. 中国食品药品检定研究院化学药品检定所,北京 102629
  • 收稿日期:2023-06-28 出版日期:2024-07-15 发布日期:2024-07-31
  • 通讯作者: * 陈华,男,硕士,主任药师,药物分析与药品质量监管。E-mail: chenhua@nifdc.org.cn;#为共同通信作者。
  • 作者简介:胡笑文,男,博士,主管药师,药物信息学。
  • 基金资助:
    国家自然科学基金资助项目（82104202）

Improving DILI prediction methods based on defined daily dose

HU Xiaowen, ZHANG Caiyu, WANG Fengfeng, PU Hengting, LIU Yang#, CHEN Hua*   

  1. Institute for Control of Chemical Drugs, National Institutes for Food and Drug Control, Beijing 102629, China
  • Received:2023-06-28 Online:2024-07-15 Published:2024-07-31

PDF (PC)

75

摘要： 目的 基于支持向量机算法,研究规定日服用剂量对肝毒性预测模型预测准确度的影响。方法 从公开数据库中收集药物肝毒性、结构和规定日服用剂量信息,得到207条数据。将数据集按4∶1分割为训练集和测试集,提取定量评估类药性理化性质作为特征,并加入规定日服用剂量作为新特征,基于支持向量机构建肝毒性预测模型,评估模型性能。对数据随机分割100次,重复上述建模步骤,考察加入新特征后,模型预测性能的变化。结果 加入规定日服用剂量后,支持向量机在测试集上的主要评估指标都有所提升,平均准确率、召回率、精准度和受试者工作曲线的AVC分别为0.763、0.773、0.779、0.832,相对于不加入新特征,分别提升了0.088、0.103、0.074、0.105。结论 规定日服用剂量能够明显提升肝毒性预测模型的预测准确性。

关键词: 肝毒性, 支持向量机, 规定日服用剂量, 预测模型, 安全性

Abstract: Objective To evaluate the impact of the defined daily dose on the performance of drug-induced liver injury (DILI) prediction models based on the support vector machine (SVM). Methods A total of 207 pieces of data on the structure and daily defined dose (DDD) were collected from public databases. The dataset was randomly split into a training set and a test set at the ratio of 4 : 1. Quantitative estimates of drug-likeness properties were extracted and the DDD was added as a new feature. The SVM was used to construct a DILI prediction model. Four metrics were used to evaluate the model performance. The dataset was randomly split 100 times to establish the predictive model, and the changes in the predictive performance of the model after DDD features were added were investigated. Results The prediction results of the SVM showed that most metrics were improved after DDD was added so that the mean accuracy, recall, precision and area under the receiver operating characteristic curve were 0.763, 0.773, 0.779 and 0.832, respectively, which were 0.088, 0.103, 0.074 and 0.105 higher than those without DDD, respectively. Conclusion The DDD can significantly improve the accuracy of the DILI prediction model.

Key words: drug induced liver injury, support vector machine, defined daily dosage, prediction model, safety

中图分类号: