[1]刘昱,任真.基于图神经网络和傅里叶变换的分子属性预测[J].中国医学物理学杂志,2026,43(3):381-385.[doi:DOI:10.3969/j.issn.1005-202X.2026.03.016]
　LIU Yu,REN Zhen.Molecular property prediction based on graph neural network and Fourier transform[J].Chinese Journal of Medical Physics,2026,43(3):381-385.[doi:DOI:10.3969/j.issn.1005-202X.2026.03.016]
点击复制

基于图神经网络和傅里叶变换的分子属性预测()

《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]

卷:
43卷
期数:
2026年第3期
页码:
381-385
栏目:
医学生物信息
出版日期:
2026-03-27

文章信息/Info

Title:
Molecular property prediction based on graph neural network and Fourier transform
文章编号:
1005-202X（2026）03-0381-05
作者:
刘昱; 任真
甘肃中医药大学医学信息工程学院， 甘肃 兰州730000
Author(s):
LIU Yu;  REN Zhen
School of Medical Information Engineering, Gansu University of Chinese Medicine, Lanzhou 730000, China
关键词:
分子属性预测; 图神经网络; 傅里叶变换
Keywords:
Keywords: molecular property prediction graph neural network Fourier transform
分类号:
R318；R34
DOI:
DOI:10.3969/j.issn.1005-202X.2026.03.016
文献标志码:
A
摘要:
当前的分子属性模型集中在预训练任务的改进，未对适用于分子属性预测的基础模型进行探索。本研究将无权图和有权图分别放入基于谱域和基于空域的图卷积神经网络，并使用傅里叶滤波器对节点特征进行去噪，最终模型在MoleculeNet的相关数据集上进行验证和消融实验。结果表明，该模型优于所有未经预训练的相关模型。与经典的图神经网络相比，该模型在分子属性预测任务中，对于分子图的适用性更强，能更高效地捕捉分子的结构和属性信息。
Abstract:
Abstract: Current molecular property models primarily focus on optimizing pre-training tasks, yet they lack exploration of foundational models suitable for molecular property prediction. Therefore, a novel model that integrates unweighted and weighted graphs into spectral-domain and spatial-domain graph convolutional neural networks is proposed, with Fourier filters employed to denoise node features. The proposed model is validated on relevant datasets from MoleculeNet, with ablation experiments conducted to confirm its effectiveness. Experimental results demonstrate that the model outperforms all non-pretrained baseline models. Compared with classical graph neural networks, the proposed model exhibits enhanced adaptability to molecular graphs and enable more efficient capture of molecular structural and property information, making it a promising approach for molecular property prediction tasks.

相似文献/References:

[1]陈浩,秦玉芳,陈明.基于图神经网络协同过滤的药物疾病关联预测[J].中国医学物理学杂志,2023,40(6):780.[doi:DOI:10.3969/j.issn.1005-202X.2023.06.018]
　CHEN Hao,QIN Yufang,et al.Collaborative filtering based on graph neural network for drug-disease association prediction[J].Chinese Journal of Medical Physics,2023,40(3):780.[doi:DOI:10.3969/j.issn.1005-202X.2023.06.018]
[2]徐信毅,李斌,朱耿,等.基于时空图卷积神经网络的精神分裂症识别[J].中国医学物理学杂志,2024,41(2):227.[doi:DOI:10.3969/j.issn.1005-202X.2024.02.016]
　XU Xinyi,LI Bin,ZHU Geng,et al.Spatial-temporal graph convolutional neural network for schizophrenia recognition[J].Chinese Journal of Medical Physics,2024,41(3):227.[doi:DOI:10.3969/j.issn.1005-202X.2024.02.016]

备注/Memo

备注/Memo:
【收稿日期】2025-04-02? 【基金项目】甘肃省自然科学基金（23JRRA1719） 【作者简介】刘昱，硕士研究生，研究方向：图神经网络、分子表征学习，E-mail: 1587221637@qq.com 【通信作者】任真，副教授，硕士生导师，研究方向：医学信息处理、深度学习，E-mail: rz@gszy.edu.cn
更新日期/Last Update: 2026-03-30