[1]方舒言,刘 宇*,侯阿龙,等.基于原子特性知识增强的分子毒性预测方法[J].计算机技术与发展,2024,34(03):155-162.[doi:10. 3969 / j. issn. 1673-629X. 2024. 03. 023]
　FANG Shu-yan,LIU Yu*,HOU A-long,et al.A Molecular Toxicity Prediction Method Based on Knowledge Enhancement of Atomic Properties[J].,2024,34(03):155-162.[doi:10. 3969 / j. issn. 1673-629X. 2024. 03. 023]

点击复制

基于原子特性知识增强的分子毒性预测方法()

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

34

期数:

2024年03期

页码:

155-162

栏目:

人工智能

出版日期:

2024-03-10

文章信息/Info

Title:

A Molecular Toxicity Prediction Method Based on Knowledge Enhancement of Atomic Properties

文章编号:

1673-629X(2024)03-0155-08

作者:

方舒言¹ ; 刘 宇¹; 2* ; 侯阿龙¹ ; 秦欢欢³ ; 刘 嵩³; 4

1. 武汉科技大学 计算机科学与技术学院,湖北 武汉 430072;

2. 湖北省智能信息处理与实时工业系统重点实验室,湖北 武汉 430072;

3. 武汉科技大学 医学院,湖北 武汉 430072;

4. 湖北省职业危害识别与控制湖北省重点实验室,湖北 武汉 430072

Author(s):

FANG Shu-yan1 ; LIU Yu1; 2* ; HOU A-long1 ; QIN Huan-huan3 ; LIU Song3; 4

1. School of Computer Science and Technology,Wuhan University of Science and Technology,Wuhan 430072,China;

2. Hubei Key Laboratory of Intelligent Information Processing and Real-time Industrial Systems,Wuhan 430072,China;

3. School of Medicine,Wuhan University of Science and Technology,Wuhan 430072,China;

4. Hubei Key Laboratory of Occupational Hazard Identification and Control in Hubei Province,Wuhan 430072,China

关键词:

分子毒性预测; 自监督学习; 知识增强; 药物发现; 摩根指纹

Keywords:

molecular toxicity prediction; self-supervised learning; knowledge enhancement; drug discovery; Morgan fingerprint

分类号:

TP183

DOI:

10. 3969 / j. issn. 1673-629X. 2024. 03. 023

摘要:

当前基于深度学习的化学分子毒性预测方法主要利用了分子的字符串表示,但现有的字符串表示模型忽视了分子中不同原子的特性知识,从而导致学习模型未能充分利用领域知识。 针对上述问题,提出了显式引入氢原子及利用摩根指纹半径增强原子特性知识的方法,使得毒性预测模型能够学习到化学分子中原子的特性知识。 在改进的毒性预测模型中,用氢原

子及原子特性知识增强的分子摩根指纹标识符序列作为输入,并在嵌入层额外引入了分子摩根指纹的半径特征。 为了验证方法的有效性,对预训练后的模型在主流的毒性预测数据

集 Tox21 上进行了微调和测试。 实验结果表明,相比于现有的基于分子序列的化学分子毒性预测方法,改进的方法在多个通道上取得了最佳的 AUC 分数。

Abstract:

Current deep learning-based methods for toxicity prediction of chemical molecules mainly utilize the string representation ofmolecules,but the existing string representation models ignore the knowledge of the properties of different atoms in molecules, whichleads to the failure of learning models fully utilizing the domain knowledge. To address?

these problems, a method that explicitlyintroduces hydrogen atoms and enhances the knowledge of atomic properties using the Morgan fingerprint radius is proposed to enable thetoxicity prediction model to learn the knowledge of the properties of atoms in chemical molecules. In the improved toxicity predictionmodel,a sequence of molecular Morgan fingerprint identifiers enhanced with hydrogen atoms and atomic property knowledge is used asinput,and the radius feature of molecular Morgan fingerprint is additionally introduced in the embedding layer. To validate theeffectiveness of the proposed method,the pre - trained model was fine - tuned and tested on the mainstream toxicity prediction datasetTox21. The experimental results showed that the improved method achieved the best AUC scores on multiple channels compared with theexisting molecular sequence-based chemical molecule toxicity prediction methods.

常用功能